[Federal Register Volume 64, Number 70 (Tuesday, April 13, 1999)]
[Proposed Rules]
[Pages 18084-18300]
From the Federal Register Online via the Government Publishing Office [www.gpo.gov]
[FR Doc No: 99-8808]
[[Page 18083]]
_______________________________________________________________________
Part II
Department of Housing and Urban Development
_______________________________________________________________________
Office of Federal Housing Enterprise Oversight
_______________________________________________________________________
12 CFR Part 1750
Risk-Based Capital; Proposed Rule
Federal Register / Vol. 64, No. 70 / Tuesday, April 13, 1999 /
Proposed Rules
[[Page 18084]]
DEPARTMENT OF HOUSING AND URBAN DEVELOPMENT
Office of Federal Housing Enterprise Oversight
12 CFR Part 1750
RIN 2550-AA02
Risk-Based Capital
AGENCY: Office of Federal Housing Enterprise Oversight, HUD.
ACTION: Notice of proposed rulemaking.
-----------------------------------------------------------------------
SUMMARY: The Office of Federal Housing Enterprise Oversight (OFHEO) is
directed by the Federal Housing Enterprises Financial Safety and
Soundness Act of 1992 to develop a risk-based capital regulation for
Freddie Mac and Fannie Mae (collectively, the Enterprises). The
regulation specifies the risk-based capital stress test that will
determine the amount of capital an Enterprise is required to hold to
maintain positive capital throughout a ten-year period of economic
stress. The results of the risk-based capital stress test will be used
to determine each Enterprise's risk-based capital requirements and,
along with the minimum capital requirement, to determine each
Enterprise's capital classification for purposes of possible
supervisory action.
This Notice of Proposed Rulemaking is the second of two notices of
proposed rulemaking pertaining to the risk-based capital regulation,
both of which respond to comments received on the Advance Notice of
Proposed Rulemaking. The first Notice of Proposed Rulemaking describes
the methodology and rationale OFHEO used to identify the proposed
benchmark loss experience, which is used to determine Enterprise credit
losses during the stress test, and proposes the use of OFHEO's House
Price Index in the stress test. The second Notice of Proposed
Rulemaking specifies the interest rate risk and other components of the
stress test, as well as the overall structure of the test.
DATES: Comments regarding this NPR must be received in writing on or
before August 11, 1999.
ADDRESSES: Send written comments to Anne E. Dewey, General Counsel,
Office of General Counsel, Office of Federal Housing Enterprise
Oversight, 1700 G Street, NW., Fourth Floor, Washington, D.C. 20552.
Written comments may also be sent by electronic mail at
RegComments@OFHEO.gov.
FOR FURTHER INFORMATION CONTACT: Patrick J. Lawler, Director of Policy
Analysis and Chief Economist; David J. Pearl, Director, Office of
Research, Analysis and Capital Standards; or Gary L. Norton, Deputy
General Counsel, Office of General Counsel, Office of Federal Housing
Enterprise Oversight, 1700 G Street, NW., Fourth Floor, Washington,
D.C. 20552, telephone (202) 414-3800 (not a toll-free number). The
telephone number for the Telecommunications Device for the Deaf is
(800) 877-8339.
SUPPLEMENTARY INFORMATION: The Supplementary Information is organized
according to this table of contents:
I. Introduction
A. Background
B. Statutory Requirements for Risk-Based Capital
C. History of the Development of the Regulation
II. Structure and Operation of the Regulation
A. Summary of the Stress Test
1. Introduction
2. Data
3. Stress Test Conditions
4. Mortgage Performance
5. Other Credit Factors
6. Cash Flows
7. Enterprise Operations & Taxes
8. Financial Reporting
9. Calculation of the Risk-based Capital Requirement
B. Sensitivity of Capital Requirement to Risk
1. MBS Guarantees (Sold Loans)
2. Commitments
3. Assets and Liabilities
4. Administrative Costs
5. External Economic Conditions
C. Implications of the Proposed Rule
1. Capital Requirements Under the Proposed Rule
2. Enterprise Adjustments to Meet the Proposed Standard
3. Guarantee Fees
4. Mortgage Interest Rates
III. Issues, Alternatives Considered
A. Mortgage Performance
1. Statutory Requirements
2. Overview of Mortgage Performance
3. Statistical Models of Mortgage Performance
4. General Methodological Issues
5. Default/Prepayment Issues
6. Loss Severity
7. Relating Losses to the Benchmark Loss Experience
8. Inflation Adjustment
B. Interest Rates
1. Yields on Treasury Securities
2. Yields of Non-Treasury Instruments
C. Mortgage Credit Enhancements
1. Background
2. Modeling Approach
3. Comments and Alternatives Considered
D. Liabilities and Derivatives
1. Modeling Methodology
2. Foreign Currency Linked or Unusual Instruments
3. Call and Cancellation Options
4. Counterparty Risk
E. Non-mortgage Investments
F. Other Housing Assets
1. Mortgage Revenue Bonds
2. Private Label REMICs
3. Interests in Partnerships and Joint Ventures
G. Commitments
1. Definition of the Term ``Commitment''
2. Retained vs. Securitized Mortgages
3. Modeling Delivery Percentages
4. Delivery Timing
5. Loan Mix Distribution
6. No New Business Rule
H. New Debt and Investment Rules
1. Rationale for New Debt and New Investment Rules
2. Analysis of ANPR Comments
I. Operating Expenses
J. Dividends and Other Capital Distributions
1. Introduction
2. Statutory Provisions
3. Proposed Approach
4. Analysis of ANPR Comments
K. Other Off-Balance Sheet Guarantees
L. Calculation of the Risk-Based Capital Requirement
1. Proposed Approach to Calculating Capital
2. Justification for Using a Present Value Approach
IV. Technical Supplement
A. Purpose and Scope
B. Single Family Default/Prepayment
1. Introduction
2. Conceptual Framework
3. Data
4. Specification of the Statistical Model
5. Explanatory Variables for Default and Prepayment
6. Empirical Results
7. Application of the Models in the Stress Test
8. Consistency with the Historical Benchmark Experience
9. References
C. Single Family Loss Severity
1. Introduction
2. Conceptual Framework
3. Data
4. Statistical Analysis
5. Consistency with the Benchmark Loss Experience
6. Application to the Stress Test
7. References
D. Multifamily Default/Prepayment
1. Introduction and Conceptual Framework
2. Historical Data
3. Statistical Estimation
4. Explanatory Variables
5. Results of the Statistical Estimation of Default and
Prepayment Equations
6. Application to the Stress Test
7. References
E. Multifamily Loss Severity
1. Introduction
2. Conceptual Framework
3. Sources of Data
4. Data Analysis
5. Application to the Stress Test
6. References
F. Property Valuation
1. Introduction
2. Conceptual Framework
3. Data Sources
4. Statistical Analysis
V. Regulatory Impact
A. Executive Order 12612, Federalism
B. Executive Order 12866, Regulatory Planning and Review
[[Page 18085]]
C. Executive Order 12988, Civil Justice Reform
D. Regulatory Flexibility Act
E. Paperwork Reduction Act
I. Introduction
A. Background
The Office of Federal Housing Enterprise Oversight (OFHEO) was
established by title XIII of the Housing and Community Development Act
of 1992, Pub. L. No. 102-550, known as the Federal Housing Enterprises
Financial Safety and Soundness Act of 1992 (1992 Act). OFHEO is an
independent office within the U.S. Department of Housing and Urban
Development (HUD) with responsibility for ensuring that the Federal
Home Loan Mortgage Corporation (Freddie Mac) and the Federal National
Mortgage Association (Fannie Mae) (collectively, the Enterprises) are
adequately capitalized and operating in a safe and sound manner.
Included among the express statutory authorities of OFHEO's Director
(the Director) is the authority to issue regulations establishing
minimum and risk-based capital standards.\1\
---------------------------------------------------------------------------
\1\ 1992 Act, section 1313(b)(1) (12 U.S.C. 4513(b)(1)).
---------------------------------------------------------------------------
Fannie Mae and Freddie Mac are Government-sponsored Enterprises
with important public purposes.\2\ These include providing liquidity to
the residential mortgage market and increasing the availability of
mortgage credit benefiting low-and moderate-income families and areas
that are underserved by lending institutions. The Enterprises engage in
two principal businesses: investing in residential mortgages and
guaranteeing securities backed by residential mortgages. The securities
the Enterprises guarantee and the debt instruments they issue are not
backed by the full faith and credit of the United States and nothing in
this document should be construed otherwise.\3\ Yet financial markets
accord the Enterprises' securities preferential treatment relative to
securities issued by potentially higher-capitalized, fully private, but
otherwise comparable firms. The market prices for Enterprise debt and
mortgage-backed securities, and the fact that the market does not
require that those securities be rated by a national rating agency,
suggest that investors perceive that the government implicitly
guarantees those securities. This perception evidently arises from the
public purposes of the Enterprises, their Congressional charters, their
potential direct access to U.S. Department of Treasury (Treasury)
funds, and the statutory exemptions of their debt and mortgage-backed
securities (MBS) from otherwise mandatory investor protection
provisions.\4\
---------------------------------------------------------------------------
\2\ 1992 Act, sections 1331-38 (12 U.S.C. 4561-67, 4562 note).
\3\ See, Federal Home Loan Mortgage Corporation Act, section
306(h)(2) (12 U.S.C. 1455(h)(2)); Federal National Mortgage
Association Charter Act, section 304(b) (12 U.S.C. 1719(b)); and
1992 Act, section 1302(4) (12 U.S.C. 4501(4)).
\4\ See, e.g., 12 U.S.C. 24 (authorizing unlimited investment by
national banks in obligations of or issued by the Enterprises); 12
U.S.C. 1455(g), 1719(d), 1723(c) (exempting securities from
oversight from Federal regulators); 15 U.S.C. 77r-1(a) (preempting
State law that would treat Enterprise securities differently from
obligations of the United States for investment purposes); 15 U.S.C.
77r-1(c) (exempting Enterprise securities from State blue sky laws).
---------------------------------------------------------------------------
Congress created OFHEO as the safety and soundness regulator of the
Enterprises to reduce their risk of failure. Although each Enterprise
at the time had experienced profitability and sustained growth,
Congress determined that there was a need for a strong and independent
regulator to promote the capital adequacy of the Enterprises. This
determination was grounded in the recognition of many factors,
including (1) the important public purpose served by the Enterprises in
the secondary market for residential mortgages, and (2) the
Enterprises' important role in providing access to mortgage credit in
central cities, rural regions, and underserved areas.
Another important factor leading to OFHEO's creation was the
recognition that the Enterprises are largely insulated from private
market discipline relative to fully private firms. This insulation
results from the apparent investor perception of an implied guarantee,
and is best exemplified by the market's acceptance of Fannie Mae
securities in the early 1980s and the Farm Credit System's securities
in the mid-1980s when these GSEs were experiencing financial
difficulties. The absence of normal market discipline on risk-taking is
a strong argument for effective government regulation, including
capital regulation.
Congress was also concerned about the serious disruptions to the
nation's housing markets that could result from an Enterprise's
failure. In introducing legislation in the House of Representatives,
then House Banking Committee Chairman Henry Gonzalez noted that--
The savings and loan crisis and the large losses incurred by the
Federal Government to resolve the crisis, raises concerns about the
scope of other potential liabilities of the United States, including
the liabilities of Fannie Mae, Freddie Mac, and the [Federal Home
Loan] banks. These entities are privately owned federally chartered
enterprises established to meet certain credit needs. Together they
have more than $800 billion in mortgage-related liabilities.\5\
---------------------------------------------------------------------------
\5\ Comments by Rep. Gonzalez upon introducing H.R. 2900, 137
Cong. Rec. H5497 (July 16, 1991).
In expressing his view that the legislation did not go far enough
to ensure the Enterprises' safety and soundness, then Ranking Minority
---------------------------------------------------------------------------
Member Jim Leach stated that--
If there is a singular lesson of the 1980's, it is that
prudential capital ratios are critical not only for providing a
cushion between an institution's liabilities and the taxpayer's
pocket book, but they ground institutional decision-making in less
risky behavior. Where there is minimal private capital at risk there
is always an inordinate incentive to bet the bank on speculative
investments or interest rate moves. And perhaps most consequently,
capital ratios determine constraints on growth. If institutions are
allowed 50 or 100 to 1 leveraging, as occurred so recently in the
thrift industry, imprudent or conflict driven decision making can
too quickly cause disproportionate growth in certain institutions,
industries and parts of the country, with the taxpayer on the line
for management stupidity, foul play or bad luck.
Fortunately, both GSEs are well run today. Fannie, in particular
has been a major market winner as the cost of funds has declined
with more restrained levels of inflation. But Congress must
understand that if interest rates had gone up rather than down in
the 1980's, Fannie Mae would be the single largest institutional
liability the U.S. government would ever have been forced to
oversee.\6\
---------------------------------------------------------------------------
\6\ Dissenting views of Rep. Leach, Government-Sponsored Housing
Enterprises Financial Safety and Soundness Act of 1991, H.R. Rep.
No. 102-206 on H.R. 2900, at 114 (1991) (House Report).
Similarly, the Senate Report \7\ stated that--
---------------------------------------------------------------------------
\7\ Federal Housing Enterprises Regulatory Reform Act of 1992,
S. Rep. No. 102-282 (1992) (Senate Report).
Past performance indicates that [the risks of an Enterprise's
failure] are not just hypothetical. While both GSEs are currently
very prosperous, HUD estimated in a 1986 report to Congress, that
Fannie Mae was insolvent on a marked-to-market basis at year-end
1978 and did not return to solvency until 1985. Its negative net
worth reached a peak of more than $20 billion in 1981, which was
roughly 20 percent of its outstanding liabilities. Its recovery owed
partly to improved management, but also, in considerable measure to
fortuitous declines in interest rates.\8\
---------------------------------------------------------------------------
\8\ S. Rep. No. 102-282, at 10 (1992).
Because of Congress' concerns, OFHEO was established as the safety
and soundness regulator of Fannie Mae and Freddie Mac. OFHEO is
responsible for conducting examinations to ensure the Enterprises'
safety and soundness and establishing and enforcing compliance with two
types of capital
[[Page 18086]]
standards required by the 1992 Act. The first is the minimum capital
standard.\9\ Using this standard, which is based on a set of leverage
ratios, OFHEO has classified each Enterprise's capital position every
quarter since OFHEO's inception. After initially using an interim
procedure, OFHEO published a rule regarding minimum capital, which
incorporates a more careful evaluation of the credit risks associated
with swaps and other off-balance sheet obligations.\10\ The resulting
standard is comparable in its construction to the risk-based capital
standards of other financial institution regulators.
---------------------------------------------------------------------------
\9\ 1992 Act, section 1362 (12 U.S.C. 4612).
\10\ 12 CFR 1750.4; see Minimum Capital, Final Rule, 61 FR
35607, July 8, 1996.
---------------------------------------------------------------------------
The second capital standard required by the 1992 Act is the risk-
based capital standard. This standard requires each Enterprise to hold
sufficient capital to survive a ten-year period characterized by
adverse credit losses and large movements in interest rates, plus an
additional amount to cover management and operations risk.\11\ The
level of capital \12\ required under this standard for an Enterprise
will reflect that Enterprise's specific risk profile at the beginning
of each quarter for which the stress test will be run.
---------------------------------------------------------------------------
\11\ 1992 Act, section 1361 (12 U.S.C. 4611).
\12\ For purposes of the risk-based capital standard, the term
``capital'' means ``total capital'' as defined under section
1303(18) of the 1992 Act (12 U.S.C. 4502(18)) to mean the sum of the
following:
(A) The core capital of the enterprise;
(B) A general allowance for foreclosure losses, which--
(i) shall include an allowance for portfolio mortgage losses, an
allowance for nonreimbursable foreclosure costs on government
claims, and an allowance for liabilities reflected on the balance
sheet for the enterprise for estimated foreclosure losses on
mortgage-backed securities; and
(ii) shall not include any reserves of the enterprise made or
held against specific assets.
(C) Any other amounts from sources of funds available to absorb
losses incurred by the enterprise, that the Director by regulation
determines are appropriate to include in determining total capital.
The term ``core capital'' is defined under section 1303(4) of
the 1992 Act (12 U.S.C. 4502(4)) to mean the sum of the following
(as determined in accordance with generally accepted accounting
principles):
(A) The par or stated value of outstanding common stock.
(B) The par or stated value of outstanding perpetual,
noncumulative preferred stock.
(C) Paid-in capital.
(D) Retained earnings.
The core capital of an enterprise shall not include any amounts
that the enterprise could be required to pay, at the option of
investors, to retire capital instruments.
---------------------------------------------------------------------------
The risk-based standard is an essential component of the safety and
soundness regulation of the Enterprises. Without the risk-based
standard, an Enterprise might adopt risk positions of sufficient
magnitude to make a capital level that just meets the minimum standard
inadequate for maintaining a safe and sound financial condition.
However, the risk-based standard cannot, by itself, ensure
sufficient capital to meet all contingencies. While the interest rate
and credit stresses that are incorporated in the stress test, as
specified by statute, are historically unprecedented, future economic
environments may be even more adverse. Additionally, the nature of
actual future stresses may differ from the precise stresses
incorporated in the model. Furthermore, the model contains factors such
as mortgage default and prepayment rates that are based on historical
experience and therefore may be less adverse than those actually
occurring in future economic environments. Similarly, the consequences
of risks other than interest rate and credit risks may also prove more
serious than the fixed proportional amount allowed for management and
operations risk.
In addition to the risk-based standard, there is a minimum capital
standard, which requires that in the absence of large measurable risks,
the Enterprise maintain a minimally acceptable level of capital.
Complementing the two capital standards are OFHEO's examination and
enforcement authorities, which provide the knowledge and authority
necessary to require prudent management practices in all environments.
All of these regulatory mechanisms operate in tandem to promote the
safety and soundness of the Enterprises.
B. Statutory Requirements for Risk-Based Capital
The 1992 Act requires that OFHEO, by regulation, establish a risk-
based capital test (known as the stress test) which, when applied to an
Enterprise, shall determine that amount of total capital for the
Enterprise that is sufficient for the Enterprise to maintain positive
capital during the stress period. The 1992 Act also provides that, in
order to meet its risk-based capital standard, each Enterprise is
required to maintain an additional 30 percent of this amount to protect
against management and operations risk.\13\
---------------------------------------------------------------------------
\13\ 1992 Act, section 1361(c)(2) (12 U.S.C. 4611(c)(2)).
---------------------------------------------------------------------------
The 1992 Act requires that the stress test subject each Enterprise
to large credit losses on mortgages it owns or guarantees. The
frequency and severity of those losses must be reasonably related to
the highest rates of default and severity of mortgage losses
experienced during a period of at least two consecutive years in
contiguous areas of the United States that together contain at least
five percent of the total U.S. population.\14\ OFHEO is required to
identify what it has characterized as the ``benchmark loss experience''
that resulted in the highest loss rate.\15\ In this context, default
and severity behavior means the frequency, timing, and severity of
losses on mortgage loans, given the specific characteristics of those
loans and the economic circumstances affecting those losses.
---------------------------------------------------------------------------
\14\ 1992 Act, section 1361(a)(1) (12 U.S.C. 4611(a)(1)).
\15\ In this document, the word ``benchmark,'' when used as an
adjective or a noun, refers to the benchmark loss experience.
---------------------------------------------------------------------------
The 1992 Act also prescribes two interest rate scenarios, one with
rates falling and the other with rates rising.\16\ The risk-based
capital amount is based on whichever scenario would require more
capital for the Enterprise. In prescribing the two scenarios, the 1992
Act describes the path of the ten-year constant maturity yield (CMT)
for each scenario and directs OFHEO to establish the yields on Treasury
instruments of other maturities in a manner reasonably related to
historical experience and judged reasonable by the Director.
---------------------------------------------------------------------------
\16\ 1992 Act, section 1361(a)(2) (12 U.S.C. 4611(a)(2)).
---------------------------------------------------------------------------
In the falling or down-rate scenario, the ten-year CMT decreases
during the first year of the stress period and then remains constant at
the lesser of (a) 600 basis points below the average yield during the
nine months preceding the stress period or (b) 60 percent of the
average yield during the three years preceding the stress period.
However, the 1992 Act limits the decrease in yield to 50 percent of the
average yield in the nine months preceding the stress period.\17\
---------------------------------------------------------------------------
\17\ 1992 Act, section 1361(a)(2)(B) (12 U.S.C. 4611(a)(2)(B)).
---------------------------------------------------------------------------
In the rising or up-rate scenario, the ten-year CMT increases
during the first year of the stress period and then remains constant at
the greater of (a) 600 basis points above the average yield during the
nine months preceding the stress period or (b) 160 percent of the
average yield during the three years preceding the stress period.
However, the 1992 Act limits the increase in yield to 175 percent of
the average yield over the nine months preceding the stress period.\18\
The 1992 Act recognizes that interest rates can affect credit risk,
specifically requiring that credit losses be adjusted for a
correspondingly higher rate of general price inflation if
[[Page 18087]]
application of the stress test produces an increase of more than 50
percent in the ten-year CMT.\19\
---------------------------------------------------------------------------
\18\ 1992 Act, section 1361(a)(2)(C) (12 U.S.C. 4611(a)(2)(C)).
\19\ 1992 Act, section 1361(a)(2)(E) (12 U.S.C. 4611(a)(2)(E)).
---------------------------------------------------------------------------
The Act requires that the stress test take into account
distinctions among mortgage product types and differences in seasoning.
It may also take into account any other factors that the Director deems
appropriate. The 1992 Act does not require a specific adjustment for
any of these factors, allowing the Director to determine how best to
account for them. Likewise, the 1992 Act requires the Director to
determine losses and gains on Enterprise activities not specifically
addressed, and all other characteristics of the stress test not
explicitly defined in the 1992 Act, on the basis of available
information, in a manner consistent with the stress test.\20\ These
stress test characteristics could include, among others, mortgage
prepayment rates and Enterprise funding activities, operating expenses,
and capital distribution activities.
---------------------------------------------------------------------------
\20\ 1992 Act, sections 1361(b) and (d)(2) (12 U.S.C. 4611(b)
and (d)(2)).
---------------------------------------------------------------------------
The 1992 Act requires the stress test to provide initially that
each Enterprise will conduct no new business within the stress period,
except to fulfill contractual commitments to purchase mortgages or
issue securities. Four years after the final risk-based capital
regulation is issued, OFHEO is authorized to modify the stress test to
incorporate assumptions about additional new business conducted during
the stress period.\21\ In doing so, OFHEO is required to take into
consideration the results of studies conducted by the Congressional
Budget Office and the Comptroller General of the United States on the
advisability and appropriate forms of new business assumptions. The
1992 Act requires that the studies be completed within the first year
after issuance of the final regulation.\22\
---------------------------------------------------------------------------
\21\ 1992 Act, sections 1361(a)(3)(B) and (D) (12 U.S.C.
4611(a)(3)(B) and (D)).
\22\ 1992 Act, section 1361(a)(3)(C) (12 U.S.C. 4611(a)(3)(C)).
---------------------------------------------------------------------------
In developing this proposal, OFHEO considered whether it would be
permissible and appropriate not to propose a detailed risk model, and
instead to rely on the risk models developed by the Enterprises
themselves.\23\ Under such a regulatory approach, OFHEO would specify
only the basic interest rate and credit assumptions, rely on the
Enterprises' internal modeling of these scenarios and review those
models and the results.
---------------------------------------------------------------------------
\23\ This approach, which OFHEO considered in detail as it began
to develop the risk-based capital regulation, was raised most
recently by Fannie Mae during the OMB review process. See the
letters from Ms. Jamie S. Gorelick, Vice Chair, Fannie Mae of
December 4, 1998 to various OMB officials; and of March 10, 1999, to
Dr. Janet Yellen, Chair, Council of Economic Advisers.
---------------------------------------------------------------------------
OFHEO has thoroughly considered this approach and believes that it
would not be consistent with the 1992 Act, which anticipates that a
publicly-available, transparent and reproducible test would be applied
to the Enterprises. The 1992 Act provides for both Enterprises to be
subject to the same stress test; \24\ that the full test be subject to
notice and comment rulemaking; \25\ that the risk-based capital
regulation be sufficiently specific to permit anyone to apply the test,
given relevant Enterprise data; \26\ and that OFHEO must make the
stress test model public.\27\ Relying on the Enterprises to compute
their own capital requirements with their proprietary models would be
inconsistent with all of these provisions.
---------------------------------------------------------------------------
\24\ See 12 U.S.C. 4611(a) (``The Director shall, by regulation,
establish a risk-based capital test for the Enterprises. When
applied to an Enterprise, the risk-based capital test shall
determine the amount of total capital for the Enterprise . . .'')
(emphasis added). See also H.R. Rep. No. 102-206 at 62 (1991).
(``Beyond these traditional capital ratios, the bill sets forth
guidelines for the creation, in highly specific regulations, of a
risk-based capital standard . . . The model, or stress test, will
generate a number for each Enterprise, which will become the risk-
based standard for that Enterprise.'') (emphasis added).
\25\ Section 1361(e)(1), 12 U.S.C. 4611(e)(1).
\26\ Section 1361(e)(2), 12 U.S.C. 4611(e)(2).
\27\ Section 1361(f), 12 U.S.C. 4611(f).
---------------------------------------------------------------------------
Moreover, a rule that specifies the details of the model will
provide a more consistent and effective capital regulation and will not
place undue burdens on the Enterprises. The structure of OFHEO's
regulatory and enforcement authorities presumes a strong risk-based
capital standard. The level of the minimum (leverage) capital standard
was established with the assumption that there would be a meaningful
risk-based standard that would address actual or potential risk not
addressed by simple leverage ratios. In addition, important OFHEO
enforcement authorities are tied to the risk-based capital requirement.
An Enterprise's failure to meet these requirements triggers two
important enforcement authorities: the ability to reduce or eliminate
the Enterprise's dividends and the ability to require a capital
restoration plan acceptable to OFHEO. Also, the grounds for a cease and
desist action vary depending on whether an Enterprise meets the risk-
based standard. Thus, a weaker standard would weaken OFHEO's
enforcement authorities.
These objectives are best obtained by a clear standard that is
presented to the public for comment and then employed consistently to
evaluate both Enterprises. Reliance instead on Enterprise models would
likely result in a weaker inconsistently-applied standard. Use of
Enterprise models would give the Enterprises broad discretion to
determine their own risk-based capital requirements because stress test
details beyond basic assumptions and modeling techniques can have a
substantial cumulative effect on the results. Existing market
distortions would give the Enterprises incentives to adjust those
details to produce low requirements.
The Enterprises' status as government-sponsored-enterprises
attenuates market discipline of Enterprise capital levels. The
Enterprises are highly leveraged financial institutions. Fully private
firms that depend heavily on debt markets are inhibited from taking on
large amounts of risk relative to their equity capital. Interest rates
on debt or guaranteed securities are sensitive to the perceived credit
quality of the issuers or guarantors. However, because investors treat
Enterprise obligations as implicitly guaranteed by the Federal
government, the normal linkage between the adequacy of an Enterprise's
capital and the interest rates on its obligations is severed. Thus,
because of the perceived implicit guarantee, the Enterprises have an
incentive to hold less capital, relative to their risk levels, than
they would if their debt costs were subject to normal market forces. A
strong risk-based capital standard can address this distortion, but the
Enterprises have little incentive to assist in producing such a result.
Reliance on different Enterprise internal models would also result
in unequal treatment. The nature of business risks and risk management
techniques are very similar at the two Enterprises. It is most
appropriate and most fair to determine each Enterprise's capital
adequacy in the same way. However, capital models developed by the two
Enterprises would likely differ significantly. Differences in resulting
standards could easily mask significant differences in true capital
adequacy between the Enterprises. Furthermore, a lower effective
standard at one Enterprise could give that Enterprise important
business advantages over the other. The resulting competitive pressures
would give the Enterprise with the higher standard an incentive to
conform with the lower standard.
A model fully specified in regulation and administered by OFHEO, on
the other hand, does not suffer these disadvantages. Such a model is
feasible
[[Page 18088]]
because OFHEO regulates only two institutions, with similar risks and
relatively narrow lines of business. The transparency of this approach
allows all interested parties to comment meaningfully on the precise
method of determining Enterprise capital requirements, and it gives the
Enterprises the ability to internalize the model for planning purposes.
In analyzing this issue, OFHEO is aware that some Federal financial
institution regulators make limited use of internal models. However,
those uses of internal models are made in very different circumstances
and by regulators with different authorizing statutes. Many of the
institutions in which these regulators rely upon internal models are
exposed to substantial market discipline of their capital and risk
positions because they rely heavily on uninsured liabilities. Such
discipline effectively forces large banks to hold capital well in
excess of regulatory requirements.
Even in these circumstances, other regulators depend on internal
models only to a small extent as a supplement to other measures of
capital adequacy. Bank capital requirements are primarily based on
overall or risk-weighted ratios that are substantially higher than
those applied to the Enterprises under the minimum capital standard. To
supplement those ratios, regulators require banks with significant
market risk exposures (those that have large trading accounts) to use
their internal value-at-risk models to calculate a market-risk capital
component of their overall risk-based capital requirements. However,
partly because of the uncertainties surrounding model construction and
verification, bank regulators require a multiple of three or more times
the amount of capital for market risk exposures that the internal
models estimate.\28\ This limited use of internal models in very
different circumstances does not appear applicable to Enterprise
capital regulation.
---------------------------------------------------------------------------
\28\ See, for example, Darryll Hendricks and Beverly Hirtle,
``Bank Capital Requirements for Market Risk: The Internal Models
Approach,'' in Economic Policy Review, Federal Reserve Bank of New
York, December 1997, pp. 3-6.
---------------------------------------------------------------------------
OFHEO considered whether an internal models approach could permit
greater flexibility and innovation by the Enterprises, because they
could modify their internal risk models at will. OFHEO believes the
issues of flexibility and innovation have been appropriately addressed
in the proposed regulation. In general, OFHEO expects that credit and
interest rate risk of new Enterprise activities and instruments will be
reflected in the stress test by simulating their credit and cash flow
characteristics using the approaches described in the regulation. OFHEO
will provide the Enterprises with its estimate of the capital treatment
of new products, investments or instruments as soon as possible after
the Enterprises notify OFHEO of the new activities. In addition, OFHEO
will monitor the Enterprises' activities and, when appropriate, propose
amendments to this regulation addressing the treatment of new
instruments and activities.
For all the reasons described, OFHEO believes that the approach
proposed in this Notice implements the requirement of the 1992 Act and
provides an appropriate means for ensuring the capital adequacy of the
Enterprises. In accordance with the requirements of the Administrative
Procedure Act, OFHEO is requesting comments on all of the issues raised
in this Notice of Proposed Rulemaking.
C. History of the Development of the Regulation
OFHEO's mission is to ensure that the Enterprises are adequately
capitalized and operating in a safe and sound manner. The principal
objective of the risk-based capital standard is to reduce the risk of
Enterprise insolvency. Another important objective of the risk-based
capital standard is to align the incentives reflected in the regulatory
capital requirement with the incentives of prudent risk management. The
ultimate goal is for the Enterprises to maintain the financial health
necessary to fulfill their public purposes. Although the stress test
produces a single capital requirement, it effectively creates
incremental regulatory capital requirements for each additional dollar
of business for every product type an Enterprise guarantees or holds in
portfolio. Marginal capital requirements for mortgages held in
portfolio will vary depending on the risk inherent in an Enterprise's
funding strategy.
OFHEO designed the stress test so that the incentives it creates
closely reflect the relative risks inherent in the Enterprises'
different activities. To this end, the proposed regulation
incorporates, to the extent feasible, consistent relationships between
the economic environment of the stress period and the Enterprises'
businesses. Doing so required OFHEO to model the Enterprises' assets,
liabilities, and off-balance sheet positions at a sufficient level of
detail to capture important risk characteristics.
However, as the level of detail of the stress test increased, so
did its complexity, along with the time and other resources that were
required to develop it. OFHEO also faced certain practical limits to
the number of variables that could be modeled due to the limitations of
existing data. Therefore, in developing this proposed regulation, OFHEO
sought to achieve a level of complexity and realism in the stress test
that appropriately balanced the associated benefits and costs.
OFHEO's stress test is comprised of a number of components, some
that correspond to subjects specifically cited in the 1992 Act and
others that represent the infrastructure that makes the stress test
operational. Figure 1 illustrates these components and their
interrelationships. The infrastructure components--database, cash
flows, and financial reports--are shaded gray. The unshaded components
implement the specific requirements of the 1992 Act, as well as the
many other aspects of the stress test that the 1992 Act either requires
or permits OFHEO to determine.
[[Page 18089]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.369
On February 8, 1995, OFHEO published an Advance Notice of Proposed
Rulemaking (ANPR) \29\ as its first step in developing the risk-based
capital regulation. The ANPR announced OFHEO's intention to develop and
publish a risk-based capital regulation and solicited public comment on
issues relating to that regulation.
---------------------------------------------------------------------------
\29\ Risk-Based Capital, ANPR, 60 FR 7468, February 8, 1995.
---------------------------------------------------------------------------
The comment period for the ANPR ended on May 9, 1995, and was
extended through June 8, 1995.\30\ OFHEO received 17 comments on the
ANPR from a variety of interested parties. Commenters included two
Executive Branch Departments, HUD and Department of Veterans Affairs
(VA); one Federal financial institution regulatory agency Office of
Thrift Supervision (OTS); one Federal regulatory agency, U.S. Commodity
Futures Trading Commission (CFTC); the Enterprises, Fannie Mae and
Freddie Mac; four trade groups, Mortgage Bankers Association of America
(MBA), America's Community Bankers (ACB), National Association of
Realtors (NAR), and Mortgage Insurance Companies of America (MICA); two
mortgage banking firms, PNC Mortgage Corporation of America and Norwest
Mortgage, Inc.), one rating agency Standard and Poor's Ratings Group
(S&P); one thrift institution, World Savings and Loan Association
(MS&L); one private mortgage research firm, Mortgage Risk Assessment
Corporation (MRAC); and one individual, Professor Anthony Yezer of
George Washington University. The responses to the ANPR ranged from a
comment on only one or two specific risk-based capital issues to an
extensive analysis of every question or issue raised. OFHEO has
considered these comments in the development of its risk-based capital
regulation.
---------------------------------------------------------------------------
\30\ Risk-Based Capital, Extension of Public Comment Period for
ANPR, 60 FR 25174, May 11, 1995.
---------------------------------------------------------------------------
OFHEO determined that the scope of the regulatory project required
the issuance of two separate Notices of Proposed Rulemaking (NPR), each
addressing different components of the stress test. On June 11, 1996,
OFHEO published a Notice of Proposed Rulemaking (NPR1),\31\ which
addresses two components. The first component is the methodology for
identifying and measuring the benchmark loss experience, which provides
the basis for determining credit losses that the Enterprises will
experience during the stress period. The second is OFHEO's proposal to
use the OFHEO House Price Index (HPI), which is a weighted repeat
transactions house price index, rather than the Constant Quality Home
Price Index (CQHPI) published by the Secretary of Commerce, to measure
differences in seasoning of single family mortgages in the stress
test.\32\ NPR1 included OFHEO's responses to all of the ANPR comments
that related to those two areas. The comment period for NPR1 ended on
September 9, 1996, and was extended through October 24, 1996.\33\ OFHEO
received 11 written comments on NPR1 and will consider and respond to
those in the final risk-based capital regulation.
---------------------------------------------------------------------------
\31\ Risk-Based Capital, NPR1, 61 FR 29592, June 11, 1996.
\32\ 61 FR 29616, June 11, 1996.
\33\ Risk-Based Capital, Extension of Public Comment Period for
NPR, 61 FR 42824, August 19, 1996.
---------------------------------------------------------------------------
This Notice of Proposed Rulemaking (NPR2) specifies and proposes
for public comment all of the remaining aspects of the risk-based
capital stress test not covered in NPR1. The notice includes an
overview of the stress test, the stress test's sensitivity to risk, the
implications of the stress test for the Enterprises, and specific
issues related to the stress test. Among the specific issues discussed
are mortgage performance (i.e., default, prepayment, and loss
severity), interest rates, new debt and new investments, commitments,
dividends and other
[[Page 18090]]
capital distributions, operating expenses, credit enhancements,
liabilities and derivatives, non-mortgage investments, and capital
calculation. The notice also includes a technical supplement that
explains the derivation of equations used in the stress test. Finally
the notice contains the regulatory text which includes the regulatory
appendix that provides the technical details of the regulation.
OFHEO believes that it is important for this proposal to receive
full public review and comment. Accordingly, OFHEO invites all
interested parties to comment on the issues raised in this NPR. OFHEO
will consider comments received, together with those received on NPR1,
in the development of the final risk-based capital regulation.
II. Structure and Operation of the Regulation
A. Summary of the Stress Test
1. Introduction
OFHEO's risk-based capital regulation is part of a larger
regulatory framework for the Enterprises that includes a minimum
capital requirement and a comprehensive examination program. The
purpose of this regulatory framework is to reduce the risk of failure
of the Enterprises by ensuring that the Enterprises are adequately
capitalized and operating safely, in accordance with the 1992 Act.
OFHEO's risk-based capital requirement differs from the minimum
capital requirement by relating the required capital to the risk in an
Enterprise's financial activities. In order to determine risk-based
capital for the Enterprises, OFHEO has been charged with creating a
stress test that simulates the effects of ten years of adverse economic
conditions on the existing assets and obligations of the Enterprises.
Both the minimum and the risk-based capital requirements work in
conjunction with OFHEO's examination program to ensure that the
Enterprises are adequately capitalized and operating safely.
In creating the proposed stress test, OFHEO had to ensure that it
met all the statutory requirements outlined in the 1992 Act and that it
accurately and appropriately captured the risks related to the business
of the Enterprises. To accomplish this, OFHEO modeled both sides of the
Enterprises' balance sheets, as well as their off-balance sheet
obligations, at the level of detail necessary to capture the risk
involved. In selecting among alternative approaches, OFHEO sought to
minimize the possibility of perverse incentives in the stress test. The
regulation was designed to ensure that stresses were appropriate in
order to promote safety and soundness and ensure the Enterprises'
ability to fulfill their important public missions.
The stress test determines, as of a point in time, how much capital
an Enterprise requires to survive the economically stressful conditions
outlined by the 1992 Act. At a minimum, the stress test would be run on
a quarterly basis. The stress test takes as inputs data on an
Enterprise's assets and obligations, operations, interest rates, and
the housing market. These data are used in econometric, financial, and
accounting models to simulate Enterprise financial performance over a
ten year period called the ``stress period.'' The stress test then
computes the amount of starting capital that would permit an Enterprise
to maintain a positive capital position throughout the stress period.
To determine the risk-based capital requirement, the 1992 Act requires
that 30 percent of this amount is added to cover management and
operations risk.
This summary provides a high level description of the stress test.
For a more detailed description, refer to the Regulation Appendix. For
explanations of the reasons for the approaches taken, refer to section
III., Issues, Alternatives Considered. For detailed information on
econometric models and historical property valuation-related indexes
used in the stress test, refer to section IV., Technical Supplement.
Throughout the summary, it may be helpful to refer to the stress test
diagram, in section I., Introduction.
2. Data
The stress test utilizes data characterizing at a point in time an
Enterprise's assets, liabilities, and off-balance sheet obligations, as
well as data on economic conditions. The Enterprises submit data to
OFHEO for mortgages, securities, and derivative contracts at the
instrument level, that is, for individual mortgages, securities, and
contracts. OFHEO obtains data on economic conditions from public
sources. All these data are referred to as ``starting position data''
for the date for which the stress test is run.
For modeling efficiency, the stress test aggregates loans into
groups of loans with common risk and cash flow characteristics (``loan
groups''). For instance, 30-year fixed-rate mortgages for single family
homes in the same geographic region, originated in the same year, with
similar interest rates and LTVs,\34\ and held in an Enterprise's
portfolio, are grouped together in one loan group. In this way, over 24
million loans are aggregated into the minimum number of loan groups
that captures important risk characteristics. These loan groups,
instead of individual loans, are then used as inputs by the mortgage
performance and cash flow components of the stress test.
---------------------------------------------------------------------------
\34\ LTV is the loan to value ratio, which is the loan balance
divided by the value of the property securing the loan.
---------------------------------------------------------------------------
In addition to starting position data for existing loans, the
stress test creates loan group data for the new mortgages that will be
added during the stress test. The 1992 Act requires that the stress
test simulate the fulfillment of the Enterprises' contractual
commitments, outstanding at the start of the stress period, to purchase
and/or securitize mortgages. The new mortgages that the stress test
adds consist of four single family loan product types: 30-year fixed-
rate, 15-year fixed-rate, adjustable-rate, and balloon. The percentage
of each type added is based on the relative proportions of those types
of loans securitized by an Enterprise that were originated during the
six months preceding the start of the stress period. The mix of LTV,
region, guarantee fee, and other characteristics of these new loans
also reflects the characteristics of the loans originated during the
preceding six months. All new mortgages are securitized. In the down-
rate scenario, 100 percent of these loans are added during the first
three months of the stress period; in the up-rate scenario, 75 percent
of these loans are added during the first six months. These loan groups
are then treated like the loan groups created for loans on the
Enterprise's books at the start of the stress period.
Because of the smaller number and greater diversity of the
Enterprises' non-mortgage financial instruments (investments and debt),
the stress test projects these cash flows at the individual instrument
level, rather than at a grouped level. Data used for these projections
include the instrument characteristics that are used to model
securities, both investment and debt, as well as derivative contracts.
3. Stress Test Conditions
a. Benchmark Loss Experience
In NPR1, OFHEO proposed the methodology for identifying the
benchmark loss experience, the stressful credit conditions which are
the basis for credit losses in the stress test. With this methodology,
OFHEO identified the worst cumulative credit losses
[[Page 18091]]
experienced by loans originated during a period of at least two
consecutive years, in contiguous states encompassing at least five
percent of the U.S. population. The performance of these loans (i.e.,
the frequency, timing and severity of their losses) and the related
interest rate and housing market environment, comprise the benchmark
loss experience.
The benchmark loss experience is based on newly originated, 30-
year, fixed-rate, first lien mortgages on owner-occupied, single family
properties. The performance of these benchmark loans was a function of
their original LTVs and other characteristics, as well as the specific
house price and interest rate paths they experienced. The stress test
applies the path of house prices from the benchmark loss experience and
the interest rate paths required by the 1992 Act. Furthermore, the
stress test simulates the performance of an Enterprise's entire
mortgage portfolio, including loans of all types, ages, and
characteristics. Primarily for these reasons, overall Enterprise
mortgage loss rates in the stress test are much lower than the loss
rates OFHEO reported in NPR1 for benchmark loans.
When the mortgage performance models are applied to benchmark
loans, using the benchmark pattern of interest rates, losses are very
close to those identified in NPR1. The remaining difference results
from the fact that OFHEO based its mortgage performance models on all
Enterprise historical loan data, not just the limited data for
benchmark loans, and that the benchmark loss experience was
particularly severe. This difference is corrected by calibrating the
single family mortgage performance models, resulting in slight upward
adjustments of default and loss severity rates, so that they are
consistent with the benchmark loss experience.
For multifamily loans, the stress test also incorporates patterns
of vacancy rates and rent growth rates that are consistent with the
benchmark loss experience. In this manner, the stress test relates the
performance of multifamily loans to the benchmark loss experience.
b. Interest Rates
Interest rates are a key component of the adverse economic
conditions of the stress test. The 1992 Act specifies two scenarios for
the ten-year Constant Maturity Treasury yield (CMT) during the stress
period. During the first year of the stress period, the ten-year CMT:
falls by the lesser of 600 basis points below the average
yield during the nine months preceding the stress period, or 60 percent
of the average yield during the three years preceding the stress
period, but in no case to a yield less than 50 percent of the average
yield during the preceding nine months (down-rate scenario); or
rises by the greater of 600 basis points above the average
yield during the nine months preceding the stress period, or 160
percent of the average yield during the three years preceding the
stress period, but in no case to a yield greater than 175 percent of
the average yield during the preceding nine months (up-rate scenario).
Changes to the ten-year CMT occur in twelve equal monthly
increments from the starting point for the ten-year CMT, which is the
average of the daily yields for the month preceding the stress period.
The ten-year CMT stays at the new level for the remainder of the stress
period.
The stress test establishes the Treasury yield curve for the stress
period in relation to the prescribed movements in the ten-year CMT. In
the down-rate scenario the yield curve is upward sloping during the
last nine years of the stress period. In the up-rate scenario the
Treasury yield curve is flat for the last nine years of the stress
period, that is, yields of other maturities are equal to that of the
ten-year CMT.
Because many different interest rates affect the Enterprises'
business performance, the ten-year CMT and the Treasury yield curve are
not the only interest rates that must be determined. For example,
current mortgage rates affect rates of refinancing of existing
mortgages; adjustable-rate mortgages periodically adjust according to
various indexes; floating rate securities (assets and liabilities) and
many rates associated with derivative contracts also adjust; and
appropriate yields must be established for new debt and investments.
Thus, the stress test requires rates and indexes other than Treasury
yields for the entire period of the stress test. Some of the key rates
that are estimated are the Federal Funds rate, London Inter-Bank
Offered Rate (LIBOR), Federal Home Loan Bank 11th District Cost of
Funds Index (COFI), and Enterprise borrowing rates. The stress test
establishes these rates and indexes by using Autoregressive Integrated
Moving Average (ARIMA) procedures--time-series estimation techniques--
to estimate their values based on historical spreads to yields on
Treasuries of comparable maturities. The procedures use historical
information to estimate values during the stress period. To reflect the
market impact of stress test economic conditions on the Enterprises'
costs of borrowing, beginning in the second year of the stress period,
50 basis points are added to the computed yields for Enterprise debt
securities.
c. Property Values
In determining the performance (rates of default, prepayment, and
of loss severity) of an Enterprise's mortgages in the stress test, the
1992 Act requires OFHEO to consider seasoning, which the stress test
captures by the use of current LTVs. The stress test calculates the
numerator of current LTV, the current loan balance, based on the unpaid
principal balance of the loan at the start of the stress period
(starting UPB) and the amortization of the loan based on product type.
Both the starting UPB and the loan product type are included in
starting position data. The stress test uses the OFHEO HPI for the
relevant Census division to track changes in property values--the
denominator of current LTV--from the time of loan origination through
to the start of the stress period. During the stress period, changes in
property values are computed by applying the pattern of house price
changes from the benchmark loss experience.
The HPI values represent average property value appreciation. In
simulating mortgage performance, the stress test also captures
variations from average house price movements, called dispersion. For
this purpose, the stress test uses the mathematical measures of
dispersion that OFHEO publishes along with the HPI.
For multifamily properties, property values are derived from
estimates of a property's net operating income and capitalization rate
multipliers. The stress test uses loan data together with rent growth
rate and vacancy rate indexes to derive estimates of net operating
income (NOI) for multifamily loans. Index values from the benchmark
loss experience are applied to starting property values to derive
current estimates of NOI for each month of the stress period. NOI is
multiplied by a capitalization rate multiplier, reflecting current
interest rates, to generate a property value. For example, if annual
NOI is $200,000 and the capitalization rate multiplier is ten, the
property value is $200,000 x 10, or $2,000,000. This value is the
denominator for current LTV for multifamily loans.
When the ten-year CMT increases by more than 50 percent over the
average yield during the nine months preceding the stress period, the
stress test takes general price inflation into consideration.
Adjustments are made to the house price and rent growth paths of the
benchmark loss experience equal to the percentage change in the ten-
year
[[Page 18092]]
CMT in excess of 50 percent.\35\ For example, if the ten-year CMT
increases by 60 percent, house price and rent growth rates increase by
ten percent. The stress test phases in this increase in equal monthly
increments during the last five years of the stress period.
---------------------------------------------------------------------------
\35\ The stress test computes the difference between the level
of the ten-year CMT in the last nine years of the stress period and
the level of the ten-year CMT if it had increased 50 percent. The
difference in yield is compounded over a nine-year period to
determine the cumulative percentage adjustment to house prices at
the end of the stress period.
---------------------------------------------------------------------------
4. Mortgage Performance
To simulate how mortgages fare during the adverse conditions of the
stress period, the stress test uses models of mortgage performance,
that project default, prepayment and loss severity rates. These models
simulate the interaction of the patterns of house prices, residential
rents, and vacancy rates of the benchmark loss experience, as well as
stress test interest rates, and mortgage risk factors, in order to
determine the performance of Enterprise loans for each month of the
stress test. As described below in further detail, the models are based
on the historical relationship of economic and mortgage risk factors to
mortgage performance, as reflected in the historical experience of the
Enterprises.
a. Loan Groups
Rather than simulating the behavior of individual loans, the models
simulate the behavior of groups of loans with common risk
characteristics. The default and prepayment models calculate the
proportion of the outstanding principal balance for each loan group
that defaults, prepays, or makes regularly scheduled loan payments in
each of the 120 months of the stress period. Single family loans are
aggregated into loan groups based on key risk and cash flow
characteristics: product type \36\ (e.g., 30-year fixed-rate, 15-year
fixed-rate, adjustable rate, balloon), original LTV, interest rate,
origination year, remittance cycle \37\ and Census division.
Multifamily loans are similarly aggregated by product type, original
LTV, origination year, interest rate, and Census region, as well as by
debt coverage ratio (DCR) \38\ and program type. Program type
distinguishes between loans purchased individually rather than as part
of a pool, and loans subject to recourse or repurchase.\39\ These
distinctions are associated with different risk characteristics.
---------------------------------------------------------------------------
\36\ The 1992 Act requires that the stress test take into
account appropriate distinctions among mortgage product types,
including single or multifamily, fixed or adjustable interest rates
and the term of the loans.
\37\ For sold loans, the remittance cycle governs the length of
time an Enterprise holds payments remitted by the seller/servicer
before passing them through to the security investor.
\38\ DCR is the ratio of property net income to debt service.
\39\ Recourse refers to the sharing of credit risk with a
seller/servicer; repurchase refers to the obligation of a seller/
servicer to repurchase 90-day delinquent loans.
---------------------------------------------------------------------------
b. Single Family Default and Prepayment
The single family models are estimated using historical data on the
performance of Enterprise loans through 1995. To simulate defaults and
prepayments, the stress test uses a 30-year fixed-rate loan model, an
adjustable-rate loan (ARM) model, and a third model for other products,
such as 15-year loans and balloon loans. Each of the three single
family models was separately estimated based on data for the relevant
product types. Each includes a calibration adjustment, so that the
results properly reflect a relationship to the benchmark loss
experience, as described earlier.
All three single family models simulate defaults and prepayments
based on values for interest rates and property values, as described
above, and variables capturing the risk characteristics of loan groups.
The variables described below are the factors used to determine the
rates of default and/or prepayment for single family loan groups:
Mortgage Age--Patterns of mortgage default and prepayment
have characteristic age profiles; defaults and prepayments increase
during the first years following loan origination, and then peak
between the fourth and seventh years.
Probability of Negative Borrower Equity--Borrowers whose
current loan balance is greater than the current value of their
mortgaged property (reflecting negative equity) are more likely to
default than those with positive equity in their properties. The
probability of negative borrower equity within a loan group is a
function of (1) house price changes (based on the HPI), and
amortization of loan principal, which together establish the average
current LTV, and (2) the dispersion of actual house price changes
around the HPI value. Thus, even when the average current LTV for a
loan group is less than one (positive equity), some percentage of the
loans will have LTVs greater than one (negative equity).
Relative Spread--This variable is an important factor in
determining whether a borrower will prepay. It reflects the value to a
borrower of the option to prepay and refinance. The stress test uses
the relative spread between the interest rate on a loan and the current
market rate on loans as a proxy for the mortgage premium value.
Burnout--The value for this variable reflects whether a
borrower has passed up earlier opportunities to refinance at favorable
interest rates. Such a borrower is less likely to prepay the current
loan and refinance, and more likely to default in the future.
Yield Curve Slope--This variable reflects the relationship
between short and long term interest rates. The shape of the yield
curve, which reflects expectations for the future levels of interest
rates, influences a borrower's decision to prepay a mortgage. Depending
on the slope of the yield curve and the type of loan a borrower may
have incentives to refinance to a fixed-rate or an adjustable-rate
mortgage.
Original LTV--The LTV at the time of mortgage origination
serves as a proxy for factors relating to the financial status of a
borrower, which can affect the borrower's future ability to make loan
payments. Higher original LTVs, which generally reflect fewer economic
resources and greater willingness to take financial risk, increase the
probability of default and lower the probability of prepayment. The
reverse is true for lower original LTVs.
Occupancy Status--The value of this variable reflects the
higher probability of default of investor-owners compared to that of
occupant-owners. The stress test applies the portfolio-wide ratio of
investor-to occupant-owners to each loan group. The single family
default and prepayment variables are listed in Table 1.
[[Page 18093]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.186
c. Multifamily Default and Prepayment
The stress test utilizes two multifamily default models and five
multifamily prepayment models to capture the behavior of loans
purchased under different programs and loans at different stages in
their life cycles. The models were estimated using historical data
through 1995 on the performance of Enterprise multifamily loans. The
stress test applies one default model to loans purchased under cash
programs (i.e., loans purchased individually), and another to loans
purchased under negotiated programs (i.e., loans purchased as part of a
pool), because the programs have different risk profiles. The
prepayment models distinguish among product types: fully-amortizing
fixed-rate, balloon, and ARM loans; those with yield maintenance
provisions (i.e., restrictions and/or penalties for prepaying a loan
during a specified period of time); and balloon loans which have
reached their stated maturity, because these distinctions affect the
probability of prepayment.
As with the models of single family mortgage performance, the
multifamily models simulate the probability of default and prepayment
based on stress test conditions and loan group risk characteristics. To
account for specific risks associated with multifamily loans, these
loans are grouped somewhat differently from single family loans. Thus,
multifamily loans are also grouped by original DCR and program type.
All of the multifamily default and prepayment models include interest
rates, rent growth rates, and vacancy rates to characterize stress test
conditions.
The following variables are factors in determining default and
prepayment rates for multifamily loan groups:
Mortgage Age--As with single family loans, the risk of
default and prepayment on multifamily loans varies over their lives.
Relative Spread--As with single family loans, this
variable reflects the value to the borrower of the option to prepay and
refinance.
Program Restructuring--This variable captures the
difference between Enterprises' management of their original
multifamily programs and current, restructured programs. That
difference affects the probability of default.
Joint Probability of Negative Equity and Negative Cash
Flow--This variable plays a role similar to that of the probability of
negative equity for single family loans. However, negative equity is
not a sufficient condition for multifamily loan default. Residential
rental property owners tend not to default unless a property's net cash
flow is negative as well. This variable captures the joint probability
of both conditions.
Balloon Maturity Risk--To reflect the added risk of
default at the balloon maturity date, this variable gives extra weight
to the joint probability of negative equity and negative cash flow in
the year before a balloon mortgage matures.
Default Type--This variable distinguishes between loans
for which the Enterprise is responsible for foreclosure and property
disposition and loans for which the seller/servicer is responsible for
repurchasing if the loan becomes 90 days delinquent.
Current LTV--This variable captures the incentive for
borrowers to refinance in order to withdraw equity from their rental
property.
Probability of Qualifying for Refinance--This variable
captures the effect on prepayments of a borrower who would not qualify
for a new loan (one that lacks an LTV of 80 percent or less and a DCR
of 120 percent or more).
Pre-balloon Refinance Incentive--This variable gives extra
weight to the relative spread in the two years prior to the balloon
maturity. This captures the additional incentive to prepay balloon
loans after the date the yield maintenance period ends, but before the
balloon maturity date.
Conventional Market Rate for Mortgages--Similar to the
single family yield curve slope variable, this variable reflects the
incentives for borrowers with ARMs to refinance into fixed-rate
mortgages.
Value of Depreciation Write-offs--This variable captures
the effect on default rates of the value to a new purchaser of the tax
benefits associated with multifamily property ownership.
Years-To-Go in the Yield Maintenance Period--This variable
captures the decreasing effect of yield maintenance provisions during
the yield maintenance period. As the cost of the provision declines in
the later years of the yield maintenance period, the disincentive to
prepay declines.
Just like the single family default and prepayment models, the
multifamily models produce, for each loan group for each month of the
stress period, default and prepayment rates which are used in the cash
flow components of the stress test. Tables 2 and 3 list the variables
included in the multifamily default and prepayment models.
[[Page 18094]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.187
[GRAPHIC] [TIFF OMITTED] TP13AP99.188
d. Loss Severity
Credit losses are determined by multiplying default rates by loss
severity rates and loan group balances. Loss severity rates are
computed as of the date of default, and are expressed as a percentage
of unpaid principal balance of the defaulting portion of a loan group.
In general, losses comprise three elements--loss of principal,
transactions costs, and funding costs. Loss of principal is the amount
of defaulting loan UPB, offset by the net proceeds of the sale
(disposition) of the foreclosed property. Transactions costs include
[[Page 18095]]
expenses related to foreclosure, property holding and disposition
expenses. Funding costs are the costs of funding non-earning assets--
first the defaulted loans, and then the foreclosed properties prior to
disposition (except in the case of sold loans, for which four months of
interest at the passthrough rate replace four months of funding costs).
For single family loans the stress test uses an econometric model
to project the net proceeds from the sale of foreclosed properties. The
model is based on historical data on defaulted Enterprise loans, and
reflects the relationship between LTV at the time of loan default
(based on a loan's original LTV, loan amortization, and house price
changes and dispersion), and proceeds of property disposition. Just as
with models of single family default and prepayment, this model
includes a calibration adjustment to make the results consistent with
the benchmark loss experience.
For multifamily loans, sale proceeds are a fixed percentage of the
defaulting UPB, based on historical experience.
For both single family and multifamily loans, transactions costs
are fixed amounts based on historical averages computed from Enterprise
data. Funding costs are captured in a discounting process described in
the following paragraph.
Foreclosure, disposition and associated costs occur over a period
of time. In order to calculate losses associated with a default as of
the time of the default, the stress test calculates loss severity rates
by discounting the different elements of loss back to the time of
default, based on stress period interest rates. The discounting process
also captures funding costs at appropriate interest rates. For single
family loans, the timing of each element is based on averages for the
benchmark loans; for multifamily loans it is based on the historical
average for the Enterprises, using data through 1995.
The calculation of loss severity rates for two types of multifamily
loans differs from the general approach. In the case of 90-day
delinquent loans that are repurchased from Enterprise security pools by
seller/servicers, rates are a fixed amount based on Enterprise
historical experience representing claims submitted by seller/servicers
for reimbursement by the Enterprise. In the case of FHA-insured loans,
the stress test reflects no losses.
The loss severity component of the stress test generates loss
severity rates for each loan group for each month of the stress period,
which are used in the cash flow components of the stress test to
calculate credit losses for the Enterprises.
5. Other Credit Factors
a. Mortgage Credit Enhancements
In many cases, at least a portion of Enterprise losses on defaulted
loans is offset by some form of credit enhancement. Credit enhancements
are contractual arrangements with third parties that reduce Enterprise
losses on defaulted loans. By including the effect of mortgage credit
enhancements, the stress test more realistically reflects Enterprise
risks related to mortgage defaults and credit losses during the stress
period.
The stress test captures many types of credit enhancements, with
differing depths and methods of coverage, for both single family and
multifamily loans. These credit enhancements include private mortgage
insurance, recourse to seller/servicers, indemnification, pool
insurance, cash accounts, spread accounts, collateral accounts, and
specific risk-sharing agreements for certain multifamily loans.
The stress test divides mortgage credit enhancements into two
categories. One category is credit enhancements that cover losses on
certain loans up to a specified percentage of the loss incurred. This
category includes private mortgage insurance, unlimited recourse,
unlimited indemnification and, for certain multifamily loans, risk-
sharing agreements. The other category includes those credit
enhancements that cover all losses on a specified set of loans, up to a
specified total amount. This category includes limited recourse,
limited indemnification, pool insurance, cash accounts, spread accounts
and collateral accounts.
The benefits of the first category of credit enhancements are
incorporated in the calculation of monthly loss severity rates. The
loss severity rate for a specific loan group is reduced based on the
credit enhancements from the first category associated with loans in
that group. The benefits of the second category of credit enhancements
are taken into account directly in the cash flow calculations. The
dollar balance of these credit enhancements is tracked and drawn down
to offset the amount of credit losses for the covered loans in a loan
group.
b. Counterparty and Other Credit Risk
In addition to mortgage credit quality, the stress test considers
the creditworthiness of companies and financial instruments to which
the Enterprises are exposed. These include most mortgage credit
enhancement counterparties (e.g., private mortgage insurance companies
and seller/servicers), privately issued and municipal securities held
as assets, derivative counterparties, and securities guaranteed for
private issuers.
For credit enhancement counterparties, securities held as assets,
and interest rate contract counterparties, the stress test reduces--or
applies ``haircuts'' to--the amounts due from these instruments or
counterparties according to their level of risk. The level of risk is
determined by public credit ratings which the stress test classifies
into four categories: AAA, AA, A and BBB. When no rating is available,
the instrument or counterparty is rated BBB. The cash flow components
of the stress test phase in the haircuts monthly in equal increments
until the total reduction listed in Table 4 is reached in the final
month of the stress period.
[[Page 18096]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.189
The stress test also applies haircuts to reflect the impact of
impairment of counterparties for derivative contracts hedging foreign
currency denominated debt. Since counterparty impairment would reduce
the effectiveness of a hedge, the stress test reflects the associated
risk by increasing the amounts owed by an Enterprise by the haircut
percentage.
c. Other Off-Balance Sheet Guarantees
In addition to guaranteeing mortgage-backed securities they issue
as part of their main business, the Enterprises occasionally provide
guarantees for other securities. The guarantees provided by the
Enterprises enhance the liquidity and appeal of these securities in the
marketplace. These securities, notably single family and multifamily
whole loan REMIC securities \40\ and mortgage tax-exempt multifamily
housing bonds, represent a small part of the Enterprises' business and
have a significant level of credit enhancement that protects the
Enterprises from losses. The performance of these securities is not
explicitly modeled in the stress test. As a proxy for the present value
of net losses on these guarantees during the stress test, the
outstanding balance of these instruments at the beginning of the stress
period is multiplied by 45 basis points. The resulting amount is
subtracted from the lowest discounted monthly capital balance when
calculating the risk-based capital requirement.
---------------------------------------------------------------------------
\40\ Real Estate Mortgage Investment Conduit (REMIC) securities
are multiclass mortgage passthrough securities. The classes of a
REMIC security can take on a wide variety of attributes with regard
to payment of principal and interest, cash flow timing
(un)certainty, and maturity, among others.
---------------------------------------------------------------------------
6. Cash Flows
For each month of the stress period, stress test cash flow
components apply projected default, prepayment, and loss severity rates
to loan group balances to produce mortgage cash flows. The cash flow
components also reduce projected mortgage losses resulting from
offsetting credit enhancements that are not accounted for in loss
severity calculations. In addition, the cash flow components calculate
cash flows for securities that the Enterprises hold as assets, or have
issued as liabilities. They generate cash flows for derivative
instruments like interest rate swaps, caps, and floors; and they apply
the haircuts to cash flows to reflect the credit risk of securities and
counterparties other than mortgage borrowers. Projected cash flows are
the principal inputs in the creation of monthly financial statements
during the stress period, which are, in turn, the basis for the
calculation of the risk-based capital requirement.
Cash flows are generated for each single family and multifamily
loan group. For retained loans, cash flows consist of scheduled
principal, prepaid principal, defaulted principal, default losses, and
interest. For sold loans, cash flows consist of credit losses,
guarantee fee income, and float income.
Because losses on sold loans are absorbed by the Enterprises and
are not passed through to security holders, no credit losses are
reflected in cash flows calculated for Enterprise-issued MBS held as
investments (including those issued by an Enterprise and later
repurchased). The credit risk is borne by the MBS issuer rather than
the MBS investor, so the credit risk on MBS has already been taken into
account in the credit risk of sold loans. Thus, cash flows for single
class Enterprise-issued MBS held as investments consist only of
principal and interest payments. Cashflows for private label securities
consist of principal and interest payments and credit losses.\41\
Principal payments are calculated by applying default and prepayment
rates that are appropriate for the loans underlying the MBS (amounts of
defaulted principal are assumed to be passed through to investors, as
well as normal amortization). Interest is computed by multiplying the
security principal balance by the coupon rate.
---------------------------------------------------------------------------
\41\ See section II. A. 5. c., Other Off-Balance Sheet
Guarantees for a description of how credit losses for private label
securities are calculated.
---------------------------------------------------------------------------
Multi-class mortgage securities such as REMICs and strips are
treated in the same manner as single class MBS. The stress test
generates cash flows for the underlying collateral, usually single
class MBS, and applies the rules of the particular multi-class security
that govern how these cash flows are directed to determine cash flows
of the specific securities held by an Enterprise. In generating cash
flows for mortgage-linked derivative contracts, where the notional
amount of the contract is based on the declining principal balance of
specified MBS, the stress test applies the terms of each contract and
tracks the appropriate declining balances. The stress test generates
cash flows for mortgage revenue bonds by treating the bonds like single
class MBS backed by 30-year, fixed-rate single family mortgages
maturing on each bond's stated maturity date.
For non-mortgage investments, outstanding debt securities and
liability-linked derivative contracts, payments of principal and
interest are calculated for each instrument based on its
[[Page 18097]]
characteristics by applying the appropriate interest rates and
principal payment rules. For asset-backed securities, one of two
collateral prepayment speeds is applied, depending on the stress test
interest rate scenario. The stress test computes cash flows for debt
securities and liability-linked derivatives according to the rules and
structure of each instrument.
7. Enterprise Operations & Taxes
The stress test simulates the income taxes, operating expenses,
issuance of new debt or purchase of new investments, exercise of
options to retire debt early or cancel derivative contracts, and
payment of dividends by the Enterprises. The stress test computes
Federal income taxes using an effective tax rate of 30 percent.
Estimated income tax is paid by the Enterprises quarterly.
An Enterprise's operating expenses decline in proportion to the
change in the size of its combined mortgage portfolio of retained and
sold loans during the stress period. The baseline level of monthly
operating expenses at the start of the stress period is equal to one-
third of operating expenses reported by the Enterprise for the quarter
preceding the stress period.
When necessary, the stress test simulates the issuance of new debt
or purchase of new investments by the Enterprises. New debt is issued
in months when there is a shortfall of cash. All debt issued during the
stress period is six-month discount notes, at Enterprise borrowing
rates projected from the estimated yield curve. Excess cash is invested
in one-month securities bearing the six-month Treasury yield.
For each month during the stress period that a security is subject
to early redemption (call) or a derivative contract is subject to
cancellation, the stress test calculates the effective remaining yield-
to-maturity \42\ of that instrument and compares it to the yield of a
replacement security, given current stress period interest rates. If
the yield is more than 50 basis points below the cost of the existing
instrument, the call or cancellation option is exercised.
---------------------------------------------------------------------------
\42\ Yields are calculated based on the outstanding principal
balances for securities and notional amounts for derivative
contracts.
---------------------------------------------------------------------------
Capital distributions are also made during the stress period. If an
Enterprise's capital exceeds the minimum capital requirement in any
quarter, dividends on preferred stock are paid, unless payment would
reduce the Enterprise's capital to an amount below the minimum
requirement. Common stock dividends are paid only in the first four
quarters of the stress period (based on an estimate of how long capital
would remain above the risk-based requirement), and only if capital
remains above the minimum capital requirement before and after the
dividends are paid. The amount paid is directly related to the earnings
trend of the Enterprise. If the trend is positive, the dividend payout
ratio is the same as the average of the four quarters preceding the
stress test. Otherwise, dividends are based on the dollar amount per
share paid in the last quarter preceding the stress test. The stress
test does not provide for any other capital distributions, such as
repurchases of common stock.
8. Financial Reporting
To the extent applicable, the stress test makes use of Generally
Accepted Accounting Principles (GAAP). The cash flows from the
financial instruments on the books of the Enterprises are the principal
basis for the creation of pro forma financial statements that capture
an Enterprise's performance over the stress period. In addition, the
stress test accounts for numerous non-cash items on the Enterprises'
balance sheets, such as receivables and unamortized and deferred
balances. The balance sheets show the monthly total capital amount for
each Enterprise, which is used in the final calculation of risk-based
capital.
9. Calculation of the Risk-based Capital Requirement
The stress test determines the amount of capital that an Enterprise
must hold at the start date in order to maintain positive capital
throughout the ten-year stress period (stress test capital). Once
stress test capital has been calculated, an additional 30 percent of
that amount is added to protect against management and operations risk.
This total is the risk-based capital requirement.
Using the financial statements generated by the stress test, the
capital balance for each month is discounted back to the start of the
stress period. This is done for both the up-rate and down-rate
scenarios. The lowest discounted monthly capital balance is then
decreased as described above to account for securities that are
guaranteed by the Enterprises which are not explicitly modeled (other
off-balance sheet guarantees). This lowest discounted monthly balance,
if positive, represents a surplus of initial capital, that is, capital
that was not ``used'' during the stress period. If negative, it
represents a deficit of initial capital. The lowest discounted monthly
balance is then subtracted from the Enterprise's initial capital. The
resulting amount is the smallest amount of starting capital required to
maintain positive capital throughout the stress period.
For example, if an Enterprise holds starting capital of $10 billion
and the lowest discounted monthly balance is $1 billion (representing a
positive capital balance even in the worst month of the stress period),
then the amount of starting capital necessary to maintain positive
capital throughout the stress period is $9.0 billion. If the lowest
discounted monthly balance is -$1 billion (representing a negative
capital balance in the worst month), the necessary starting capital is
$11.0 billion.
In the final step, necessary starting capital is multiplied by 1.3
to complete the calculation of the risk-based capital requirement
required by the 1992 Act.
B. Sensitivity of Capital Requirement to Risk
An Enterprise's risk-based capital requirement under this proposed
regulation is sensitive to a wide variety of factors that affect
Enterprise risk. The existing minimum capital requirement depends
almost entirely on the size of an Enterprise's two principal
businesses: MBS guarantees and leveraged investments in mortgages and
in MBS. In contrast, the risk-based capital requirement depends not
only on the outstanding volumes of an Enterprise's guarantees and
assets, but also on the degree of risk taken on by the Enterprise in
connection with these businesses. Thus, the risk-based requirement is
sensitive to the characteristics of mortgages and mortgage guarantees
that affect risk, credit enhancements for those mortgages, the asset/
liability risk management strategies of the Enterprise, the value of
properties collateralizing the mortgages, and recent interest rate
levels.
In designing the stress test on which the risk-based capital
requirement is based, OFHEO sought to incorporate all significant
sources of credit and interest rate risk. OFHEO further sought to
design the stress test so that differences in specific risk factors
affect the risk-based capital requirement in amounts commensurate with
the difference in risk. To quantify the marginal effects of changes in
risk on the capital required for each scenario (required capital),
OFHEO conducted a number of sensitivity tests. OFHEO first computed the
risk-based capital requirement for each Enterprise in each interest
rate
[[Page 18098]]
scenario for June 30, 1997.\43\ These results serve as a base case.
OFHEO then made a series of small adjustments to each Enterprise's risk
positions and compared the results for all four Enterprise-scenario
combinations with the relevant base case results. The differences in
results provide a measure of the incremental changes in required
capital (which may be positive or negative) caused by the risk
adjustment.
---------------------------------------------------------------------------
\43\ The results are discussed in section II. C., Implications
of the Proposed Rule.
---------------------------------------------------------------------------
Section II. B.1., MBS Guarantees (Sold Loans), below presents the
results of sensitivity tests related to an Enterprise's guarantee
business. In each test, OFHEO simulated the effects on required capital
of a hypothetical addition to each Enterprise's outstanding MBS
guarantees (sold loans). The simulation results show, in both an
absolute and relative sense, how different characteristics of sold
loans affect required capital. Section II. B. 2., Commitments,
illustrates how required capital would be affected if each Enterprise
had had a larger volume of outstanding commitments. Section II. B. 3.,
Assets and Liabilities, discusses the effects of hypothetical additions
of retained loans accompanied by additions of debt. Section II. B. 4.,
Administrative Costs, discusses how risk-based capital would be
affected by higher administrative (operating) expenses. Finally,
Section II. B. 5., External Economic Conditions, discusses how risk-
based capital would be affected had house prices or interest rates
behaved differently than they actually did in the period just preceding
the starting date of the stress test.
Sensitivity test results differ between the two Enterprises for two
reasons. First, the risk adjustments made to the two Enterprises'
positions were not precisely the same. For example, in sensitivity
tests involving changes in outstanding sold loan volumes, each
Enterprise's additional sold loans reflect that Enterprise's typical
security remittance cycles, and remittance cycles affect the risk
characteristics of sold loans. Second, the incremental effects on
required capital of any change in an Enterprises's risk positions are
affected by the Enterprise's individual circumstances and policies. Two
examples are the Enterprise's projected Federal income tax situation
during the stress period and its dividend policies. During portions of
the stress period in which an Enterprise is paying taxes or receiving
refunds, financial gains and losses are shared with the government
because changes in income cause changes in taxes. Conversely, during
portions of the stress period in which an Enterprise has exhausted tax
carrybacks, the full benefit or cost of a change in income is
experienced by the Enterprise. In the base case, both Enterprises
exhaust their tax carrybacks mid-way through the stress period in the
down-rate scenario. In the up-rate scenarios, Fannie Mae does the same,
but Freddie Mac either pays taxes or receives refunds throughout the
stress period. An Enterprise's tax situation during the stress period
depends primarily on the Enterprise's risk exposures. The longer an
Enterprise continues to be profitable in the stress environment, the
longer it is affected by taxes.
Differences in recent dividend policies can cause small differences
in the incremental capital associated with specific changes in risk
because common stock dividends during the first year of the stress
period depend on recent dividend payouts. Differences in dividend
policies, therefore, can lead to differences in the amount of earnings
changes that are shared with stockholders.
Results are shown for both interest rate scenarios, even though
only one (the one that results in the highest required capital) can be
binding at any specific time. For June 1997, the up-rate scenario
resulted in higher required capital for Fannie Mae, while the down-rate
scenario was more adverse for Freddie Mac. However, the relative
adversity of the two scenarios may change over time for either
Enterprise depending on business strategies and market conditions.
In the tables of this section, the phrase ``incremental capital''
is used to mean the change in the amount of required capital in a
particular scenario accompanying a small change in the overall risk
profile of an Enterprise. Several considerations affect appropriate
interpretation of these numbers. First, the incremental capital
percentages shown in the tables are not fixed. As discussed below in
section II. B. 5. c., Sensitivity to Risk Characteristics in Different
Economic Environments, future business strategies and economic
conditions may alter the required capital sensitivities from those of
June 1997, which are presented here. Furthermore, bigger or smaller
changes in risk may not have a proportional effect on capital. A $20
billion increase in a particular group of loan guarantees may not have
exactly twice the effect on required capital as a $10 billion increase
in the same group of guarantees.
Second, in anticipating the effect on required capital of a change
in any risk factor, an Enterprise likely will be concerned not only
with the immediate effect, but also with the longer term effect. For
example, in considering the capital implications of making additional
mortgage guarantees, the incremental effects on required capital of the
guarantees at all future dates that the loans continue to be
outstanding are relevant. In this case, an important consideration is
that the incremental effects of mortgage guarantees generally diminish
over time.
Third, the incremental capital percentages do not determine an
amount of capital that must be added in order to accept a specific
increase in risk. As discussed below in Section II. C. 2., Enterprise
Adjustments to Meet the Proposed Standard, it may often be less costly
to increase hedges of other risks than to raise equity funds in
response to an increase in risks.
1. MBS Guarantees (Sold Loans)
The Enterprises have two principal lines of business. They function
both as guarantors of mortgage-backed securities and as leveraged
investors in mortgages and mortgage-backed securities. As guarantors,
the Enterprises receive principal and interest payments on home
mortgages, which they pass through to security investors, minus a share
of the interest payments, which they retain as a guarantee fee. Because
of differences in the timing of their receipt of funds and payments to
investors, they also earn float income (which may be positive or
negative). In return, they bear the risk of loss if a borrower
defaults, and they incur additional administrative expenses.
The stress test projects the flows of income and expenses
associated with loan guarantees based on the characteristics of the
mortgages and the economic circumstances of the stress period. The
resulting net cash inflows or outflows are directly reflected in the
Enterprise's borrowing or investing volumes during the stress period.
The interest paid or received on the new debt issues or investments
that are attributable to the guarantees have further effects on income,
borrowing, and investing volumes. Income, in turn, affects taxes,
dividends, capital, and (ultimately) required capital.
OFHEO examined the implications for required capital of risk
factors associated with sold loans as follows. After computing the
capital required under this proposed rule for data reflecting the
Enterprises' books of business and the accompanying economic
circumstances as of June 30, 1997, OFHEO added a quantity ($10 billion)
of sold loans that embodied the specific risk characteristics under
[[Page 18099]]
examination. The capital required for each scenario was then recomputed
and compared with the capital required for the same scenario before
loans were added. The difference is the incremental capital required
for the additional sold loans in that scenario. The results are
expressed as a percent of the volume of sold loans added.
Additional sold loans would normally be accompanied by additional
administrative expenses. In computing required capital for books-of-
business that included additional sold loans, OFHEO estimated the
additional costs by increasing administrative expense for each
Enterprise in proportion to the increase in that Enterprise's overall
(retained plus sold loan) portfolio. Those costs amounted to about six
basis points (0.06 percent) per year on the new sold loans for each
Enterprise. Different assumptions about administrative costs would
affect the results; Section II. B. 4., Administrative Costs, discusses
the effects on required capital of differences in administrative costs.
Section II. B. 1. a., Loans with Mixed Characteristics Reflecting
Enterprise Portfolios, discusses a simulation incorporating a general
increase in sold loans embodying the same mix of characteristics as
that found in each Enterprise's sold loan portfolio in June 1997 and
describes how the increase affects various types of income and expense
over the course of the stress period. Section II. B. 1. b., Loans with
Specific Identical Characteristics, discusses a series of simulations,
each incorporating an increase in sold loans with specific
characteristics.
a. Loans with Mixed Characteristics Reflecting Enterprise Portfolios
The first simulation (Simulation 1) was designed to examine the
incremental effects of a general increase in each Enterprise's sold
loan portfolio (MBS guarantees). The volume of each loan group
(comprising loans with a common set of risk factors) in each
Enterprise's sold loan portfolio as of June 1997 was increased
proportionally by a factor that resulted in a total of $10 billion of
additional sold loans. The results indicate the effects on risk-based
capital of a general expansion of an Enterprise's MBS guarantee
business. Alternatively, they can be viewed as the average effect on
required capital of sold loans, weighted by each Enterprise's mix of
outstanding sold loan business in June 1997. The results, expressed as
a percent of the increase in sold loans, are summarized in Table 5.
[GRAPHIC] [TIFF OMITTED] TP13AP99.190
In the up-rate scenario, a general increase in sold loans has only
a small effect on required capital for either Enterprise. For Freddie
Mac, sold loans are, on balance, a small source of strength. That is,
income generated over the course of the stress period by sold loans
(principally guarantee fees and float) exceeds related expenses
(principally loan losses and administrative expense). The reverse is
true for Fannie Mae. In the down-rate scenario, the incremental capital
required for these sold loan mixes is near 0.85 percent of the increase
in guarantees for both Enterprises. On average, the results for the two
scenarios are similar to the existing minimum capital ratios for sold
loans of 0.45 percent.
Table 6 illustrates the effects on specific income and expense
categories of the additional sold loans in Simulation 1, and how these
effects translate into changes in capital requirements.
[[Page 18100]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.191
Guarantee fees and administrative expense depend on the volume of
loans outstanding. Thus, they are sensitive to the projected
liquidation rates (the sum of prepayment, default, and amortization
rates) of the additional sold loans. In the down-rate scenario (with a
ten-year constant maturity treasury yield of 3.2 percent during the
last nine years of the stress period), loans prepay rapidly, while in
the up-rate scenario (with all treasury yields at 11.4 percent), loans
prepay slowly. As a result, in the up-rate scenario, guarantee fee
income and administrative expense are roughly 2\2/3\ times as great as
they are in the down-rate scenario.
Credit losses (charge-offs) depend on the credit risk
characteristics of the additional sold loans. They are also larger in
the up-rate scenario than in the down-rate scenario because loans
remain outstanding longer, and therefore, at risk of default. Loss
severity rates also are higher in the up-rate scenario because the
interest carrying cost on foreclosed real estate is higher. These
differences between the two scenarios are moderated by somewhat more
favorable house price behavior and by better average loan quality when
interest rates are high. Loan quality is poorer when interest rates are
low because the better quality loans are projected to prepay much
faster. Because of these offsetting influences, credit losses in the
up-rate scenario are only 1\1/3\ times as great as they are in the
down-rate scenario. Freddie Mac's credit losses are about ten percent
lower than Fannie Mae's, reflecting a slightly less risky mix of loan
characteristics.
Float income depends on security remittance cycles, interest rates,
and loan liquidation rates. This source of income on the additional
sold loans is higher, for both Enterprises, in the scenario with higher
interest rates because of lower liquidation rates and higher earnings
ratios on positive float balances. The difference is much more
pronounced for Freddie Mac because of differences in security
remittance cycles. Freddie Mac holds prepayment funds for a longer
period than Fannie Mae, earning a market rate of interest during the
extra time, while accruing liabilities to investors at the security
coupon rate. When interest rates rise, that provides extra income, but
when rates fall, net losses accrue.
Net interest income is affected because net cash inflows and
outflows associated with the other income and expense categories lead
to changes in borrowing or investing. The effects are small in the up-
rate scenario because the net flows caused by other factors are small.
The effects also are small in the down-rate scenario, even though the
net cash flows are much larger, because the interest rates associated
with new borrowing or investing are low.
Taxes reduce the effects of all income changes by 30 percent as
long as an Enterprise is paying taxes or receiving tax refunds. Because
both Enterprises, in the decreasing interest rate environment, and
Fannie Mae, in the increasing rate environment, exhaust their tax
carrybacks mid-way through the stress period, the tax effects vary
depending on the timing of income flows during the stress period.
Freddie Mac, however, performs well in the up-rate scenario, given its
June 1997 risk positions, and pays taxes or receives refunds throughout
the stress period.
[[Page 18101]]
Dividends on common stock can be affected by additional sold loans
only through changes in income during the first year of the stress
period because the stress test specifies that common stock dividends
are paid only during that year. Common stock dividends are little
affected in this simulation because income changes during the first
year are small and because dividends in the base case simulations for
Fannie Mae in both scenarios, and Freddie Mac in the down-rate
scenario, are insensitive to income. In those cases, dividends are set
at their absolute level in the quarter preceding the stress test
because of income declines during the first year. Preferred stock
dividends are unaffected in this simulation because the changes in
capital are insufficient to affect whether either Enterprise meets its
minimum capital requirement during the stress period.
The total change in capital is the sum (using the appropriate
signs) of the effects measured through all of the above income and
expense categories. The sum equals the net decline in capital at the
end of the stress period (as a percent of the increase in sold loans).
The capital position in the final month of the stress period is the
lowest during the stress period for both Enterprises in both scenarios
for the June 1997 base case, so it is the basis for the required
capital calculations in all of the simulations discussed in this
section.
The cumulative discount factor is based on after-tax borrowing or
investing interest rates. Thus, discount factors are relatively high in
the up-rate scenario. Freddie Mac's discount factor is lower than
Fannie Mae's in that scenario because taxes reduce Freddie Mac's after-
tax interest rates in the second half of the stress period, but do not
reduce Fannie Mae's. The discounted total shows the effects of the
additional sold loans on the amount of capital needed to survive the
stress test. This amount, when multiplied by 1.3 to include the
additional amount for management and operations risks, shows the
effects on required capital of the additional sold loans.
b. Loans with Specific Identical Characteristics
Unlike the first simulation, which showed the combined effects of
each Enterprise's existing mix of risk factors, the following
simulations focus on the effects of changes in specific risk factors.
In each of the following cases, the sold portfolio is increased as
before, but all of the additional loans are identical. The results show
how much required capital would be affected by additional sold loans
with specific risk characteristics and guarantee fees or,
alternatively, how much loans with such characteristics and fees
contribute to required capital. The assumptions about guarantee fees
have a significant effect on the results. Guarantee fees are generally
the same in most of these simulations in order to focus the results on
the incremental capital effects of specific risk factors. In practice,
though, the Enterprises typically vary the guarantee fees charged to a
loan seller depending on the mix of loans they receive from that
seller. Thus, the Enterprises implicitly charge higher fees for riskier
loans. It would be misleading to characterize these simulation results,
which are based on constant guarantee fees, as indicating the relative
capital implications of loans in different risk groups as typically
acquired by the Enterprises, without making an appropriate adjustment
for typical differences in effective guarantee fees. Making such an
adjustment in the model would be difficult, however, because the
Enterprises do not generally make explicit differences in guarantee
fees for individual loans with differences in risk. The same guarantee
fee typically applies to all loans in a pool of loans and may be
affected by the mix of loans in the pool.
Also, Enterprise guarantee fees remain constant over the life of
the loan, but the risk of the loan generally declines as the loan
seasons. A majority of the simulations in this subsection involve new
loans. The comparative results of such simulations provide a measure of
the relative effects on required capital of different risk factors, but
these results do not, by themselves, indicate the expected effects on
required capital of the loans over their lifetimes. Additional
simulations show the effects of loan seasoning on required capital.
In these simulations, securities were assumed to have been sold at
par with coupons equal to the contract interest rates, less the
servicing and guarantee margins. Servicing margins are 30 basis points.
For Fannie Mae, the loans were assumed to be securitized under their
standard programs with seven days of float on passthrough payments. For
Freddie Mac, their ``45-day'' security rules were assumed in float
calculations. These securities have negative three days of float on
scheduled principal and interest (payments are made to investors before
payments are received from servicers) and an average of 38 days of
float on prepayments. (In Simulation 1, both 45-day and 75-day rules
were used for Freddie Mac, based on the mix of securities outstanding
in June 1997.)
(i) Differences in Guarantee Fees
To illustrate the effect on required capital of guarantee fees, two
simulations were performed that were identical except for guarantee
fees. In Simulations 2 and 3, shown in Table 7, the additional sold
loans were all newly originated, fixed-rate mortgages (FRMs) in the
West South Central Census Division (Texas, Oklahoma, Louisiana, and
Arkansas); with 30-year terms, 7.5 percent contract interest rates, and
80 percent loan-to-value ratios (LTVs). In Simulation 2, guarantee fees
were set at 23 basis points, which is roughly the overall average rate
for the two Enterprises, but not necessarily for loans with these
characteristics. This simulation is used as a reference for comparison
in Tables 8, 11, 12, 16, 17, 19, and 20. The average rate was used in
most of the simulations involving additional single family loans for
convenience and to isolate the differential effects of other risk
factors. In Simulation 3, however, the guarantee fee was reduced to 18
basis points to isolate the effects of different guarantee fees. The
differences in the results for Simulations 2 and 3 can be used to
roughly estimate how the results of other simulations might have been
affected by other guarantee fee assumptions.
[[Page 18102]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.192
The incremental capital needed for loans in both of these
simulations is substantially higher than that needed for loans with the
mix of characteristics in Simulation 1. This result occurs mainly
because new 30-year FRMs have nearly double the credit losses in the
up-rate scenario and 50 percent more in the down-rate scenario. For
Freddie Mac, an additional reason is that securities with the 45-day
remittance cycle assumed in Simulations 2 and 3 produce substantially
less float income in the up-rate scenario and more negative float
income in the down-rate scenario than the average guarantee mix in
Simulation 1 did. Freddie Mac's capital need in the up-rate scenario is
reduced relative to Fannie Mae's because of tax effects in the second
half of the stress period.
The effect of lower guarantee fees is to increase required capital
in both scenarios. A five basis-point reduction in guarantee fees
raises required capital by 14 to 18 basis points in the down-rate
scenario. The difference in incremental capital is twice that amount in
the up-rate scenario because the loans survive longer, owing to
significantly fewer prepayments, and so the change in the fee rate
applies to a larger volume of outstanding loans during the stress
period.
(ii) Differences in Loan Age, With Slow and Steady House Price
Inflation
Seasoned loans (those not recently originated) have different risk
characteristics than new loans because loans have different
propensities to default and prepay at different ages and because the
houses collateralizing seasoned loans have experienced changes in
value. Changes in house value alter the probability of negative
borrower equity, a key factor influencing default behavior.
In Table 8, the results of Simulations 4-7, along with Simulation
2, which is repeated here, show the effects of age on risk for loans
originated in the West South Central Census Division. Houses in that
area of the country generally have experienced price appreciation near
the national average in recent years. Average annual appreciation over
the eight years ending in the second quarter of 1997 was 3.0 percent.
Table 9 shows the cumulative average appreciation for houses
collateralizing loans of different ages.
[GRAPHIC] [TIFF OMITTED] TP13AP99.193
All of the simulations reported in Table 8 are identical, except
for the age of the sold loans underlying the additional guarantees.
Given the steady increase in house prices preceding the starting point
of the simulations, loans are less likely to default over the course of
the stress period the older they are at the beginning of the period.
Cumulative credit losses for loans made eight years before the start of
the stress period are only about \1/5\ as great as for new loans in the
up-rate scenario, and about \2/5\ as great in the down-rate scenario.
In addition, loans made more than four years earlier have lower
liquidation rates than new loans, providing a larger stream of
guarantee fees. Consequently, guarantees of older loans cause much
smaller increases in capital requirements in the down-rate scenario and
actually reduce capital required in the up-rate scenario.
[[Page 18103]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.194
(iii) Differences in Past House Price Appreciation
The benefits of loan age in reducing risk can be substantially
increased or reversed by differences in house price appreciation. Table
10 shows results for simulations on four-and eight-year-old loans from
different geographic areas. Simulations 8 and 9 are the same as
Simulation 5, except the loans in Simulation 8 were made on properties
in the Mountain Census Division, where house values rose sharply after
the loans were originated, and loans in Simulation 9 were made in the
Pacific Census Division, where house values were stagnant. Similarly,
Simulations 10 and 11 are the same as Simulation 7, except for the
Census division.
[GRAPHIC] [TIFF OMITTED] TP13AP99.195
For four-year-old loans, differences in credit losses are
substantial and account for almost all differences in results. In both
scenarios, credit losses are more than 2\1/2\ times as great in the
Pacific Census Division as they are in the Mountain Census Division.
However, the effects of different previous changes in house prices
ultimately diminish. For eight-year old loans, charge-offs are only
about \1/3\ higher in the Pacific Census Division, despite increasing
disparity in house price appreciation. Furthermore, that smaller
proportional increase in charge-offs is applied to a smaller base
because charge-offs are much lower for eight-year old loans than for
four-year old loans in all three Census divisions.
(iv) Differences in Loan Age and Loan-to-Value Ratio
The higher the original loan-to-value ratio of a loan, the lower
the borrower equity. Thus, the more likely it is to default and less
likely it is to prepay. The effects of differences in original LTV,
however, generally diminish with age. Table 11 shows the results for
different LTV-age combinations for 30-year FRMs in the West South
Central Division.
[[Page 18104]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.196
In these simulations, the 95 percent LTV loans are assumed to be
covered by private mortgage insurance with 30 percent coverage, the
current Enterprise standard, provided by a double-A rated firm. Even
with the insurance coverage, however, high LTV loans are much riskier
than low LTV loans. Not only are high LTV loans more likely to default
at any time during the stress period, but they are also less likely to
prepay, especially in the down-rate scenario. Thus, they are exposed to
default risk over a longer amount of time.
For newly originated loans, the results are particularly striking.
In the up-rate scenario, credit losses on 95 percent LTV loans are very
much higher than they are for 50 percent LTV loans. In the down-rate
scenario, the difference is even greater. These differences in
performance between high and low LTV loans are much bigger than would
be expected in normal times. But the very poor credit conditions in the
stress test environment have a disproportionate effect on the more
vulnerable high LTV loans.
For seasoned loans, the effects of LTV are muted. Seasoned loans
with 50 percent LTVs reduce required capital less than comparable new
loans. Though credit losses are lower than those of newly originated
loans, the difference is minor, as credit losses are very low in both
cases. More importantly, the older loans amortize faster, reducing
guarantee fees significantly. For loans with 95 percent LTVs, the
difference in credit losses between seasoned and new loans is
substantial. With a 13.7 percent average house price appreciation since
origination, these seasoned 95 percent LTV loans perform only a little
bit worse than newly originated 80 percent LTV loans.
(v) Differences in Product Type and LTV Ratio
The simulations shown in Table 12 show the relative effects of
three different product types (30-year FRMs, 15-year FRMs, and
adjustable-rate mortgages) with low, medium, and high LTVs). All are
newly originated loans. To isolate the effects of loan type, the 7.5
percent contract loan rate was retained for the 15-year FRMs and is the
initial rate on the adjustable-rate mortgages (ARMs). The ARMs adjust
annually to 2.75 percentage points above the one-year constant maturity
Treasury yield, with a two percentage point annual adjustment cap and a
five percentage point lifetime cap.
[[Page 18105]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.197
The intermediate-term (15-year) FRMs have consistently lower credit
losses than long-term (30-year) FRMs because the shorter-term loans
amortize more quickly, and borrowers choosing those loans tend to have
greater financial resources. For 50 percent LTV loans, the difference
in credit losses is small, as credit losses are very low for loans of
both terms. In the up-rate scenario, the 30-year loans benefit from
slower amortization, which results in more guarantee fees. In both the
80 percent and 95 percent LTV categories, the more favorable
incremental capital effects of 15-year loans reflect their greater
safety. For 95 percent LTV loans, the 15-year loans have sharply lower
credit losses, nearly 90 percent below those of 30-year FRMs.
ARM loans are riskier than 30-year FRMs at all LTV levels in the
up-rate scenario, with the differences becoming more pronounced as LTV
ratios rise. ARM credit losses in the up-rate scenario are only
modestly higher than 30-year FRM credit losses for low LTV loans, but
rise to more than double those for 30-year FRMs for high LTV loans.
Credit losses for high LTV ARMs cumulate over the course of the stress
period to 13.5 percent of the initial loan balances. As the loan
interest rates adjust to their lifetime caps, some borrowers have
difficulty meeting the elevated payments.
When interest rates decline, ARMs perform much better. They prepay
much more slowly than FRMs in this environment and, therefore, produce
substantially more guarantee fee income. At low and moderate LTVs, ARMs
have more favorable capital effects than FRMs. However, the greater
sensitivity of defaults on ARMs with high initial LTVs outweighs the
benefits of higher fee income generated by such loans. While credit
losses for high LTV ARMs are still much lower in the down-rate scenario
than in the up-rate scenario, the discounted values of those losses are
larger in the down-rate scenario because the discount rates are so much
lower in that scenario. The capital effects depend on the discounted
values, so they are nearly as large in the down-rate scenario for high
LTV ARMs as they are in the up-rate scenario. Because of the high risk
associated with high LTV ARMs, the Enterprises generally have not
purchased ARMs with LTV ratios above 90 percent under their regular
underwriting guidelines.
(vi) Differences in Multifamily Loans
The Enterprises deal in a large variety of multifamily loan
products, and the products differ significantly between the
Enterprises. The simulations reported in Table 13 show the incremental
effects on required capital of multifamily loans with some relatively
common characteristics. The additional sold loans in Simulation 22 are
newly originated 15-year balloons with 70 percent LTVs, debt coverage
ratios (DCR) of 1.3.\44\ The Fannie Mae loans are assumed to provide
partial recourse to the seller for losses, while the Freddie Mac loans
do not. Accordingly, a higher guarantee fee is assumed for Freddie Mac
loans, 75 basis points, than for Fannie Mae loans, 50 basis points.
Simulations 23, 24, and 25 differ, respectively, by changing the
balloon to five years, changing the LTV to 80 percent and the DCR to
1.2, and changing the loan age to five years.
---------------------------------------------------------------------------
\44\ All of the multifamily loans were originated in the West
Census Region with 8.5 percent coupons and servicing margins of 50
basis points.
[[Page 18106]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.198
Unlike single family loans, multifamily loans with a few years of
seasoning have substantially higher credit losses during the stress
period. Both types of loans generally have low credit losses in the
first years after origination, then rise to a peak before declining.
However, the peak loss years for multifamily loans come several years
after those for single family loans. Thus, the five-year old loans in
Simulation 25 experience more bad loss years than comparable new loans
(Simulation 22). Credit losses for high LTV, low DCR loans (Simulation
23) are also higher than comparable lower LTV, higher DCR loans because
there is a higher probability that the borrower would have an economic
incentive to default during the stress period (no equity and negative
cash flow). Five-year balloons have higher losses in the up-rate
scenario because some properties would be unable to manage the higher
interest rates that would accompany a new loan. In the down-rate
scenario, five-year balloons terminate sooner and, thus, provide less
guarantee fee income.
Multifamily loan losses are generally less than guarantee fee
income in the down-rate scenario. This is especially true for newly
originated loans because most of the loans prepay before reaching their
peak loss years. Multifamily loans also benefit in the down-rate
scenario from lower capitalization rates, which improve their estimated
LTVs.
(vii) Differences in Mortgage Insurance on High LTV Loans
By law, conventional loans purchased by the Enterprises with LTVs
greater than 80 percent require credit enhancement. Of the three types
permitted, private mortgage insurance is by far the most commonly used.
As described above, simulations involving additional guarantees for
loans with 95 percent LTV ratios assume that the loans carry 30 percent
coverage by a AA rated firm. The simulations reported in Table 14 show
effects of varying insurance characteristics on single family loans.
The guarantee additions in each case are for newly originated, long-
term FRMs.
[GRAPHIC] [TIFF OMITTED] TP13AP99.199
In 1995, both Enterprises raised their coverage requirements on 95
percent LTV loans from 25 percent to 30 percent. Credit losses in
Simulation 26, with lower coverage than in Simulation 13 (but with all
other characteristics are the same), are 15 percent higher in the down-
rate scenario and 12 percent higher in the up-rate scenario than they
are in Simulation 13. Because the discounted value of those changes is
higher in the down-rate scenario, the
[[Page 18107]]
required capital is affected more significantly in that scenario.
Reducing the credit quality of the coverage (Simulation 28) has much
the same effect as reducing the amount of coverage, while improving the
credit quality (Simulation 27) has the opposite effect.
(viii) Differences in Mortgage Interest Rates
Loans with low interest rates amortize more quickly and prepay more
slowly. The reverse is true for high interest rate loans. Table 15
shows the results of simulations for newly originated, long-term FRMs
with different interest rates. In practice, loans with different
interest rates have been originated in different time periods. However,
to isolate the effects of different mortgage interest rates, all loans
are assumed to be made simultaneously.
[GRAPHIC] [TIFF OMITTED] TP13AP99.200
Faster amortization improves loan quality, so credit losses are
significantly lower for mortgages with low interest rates. Low interest
rate loans also prepay significantly more slowly in the down-rate
scenario, increasing guarantee fees. For Freddie Mac, these differences
between high and low mortgage interest rates are accentuated by
differences in float income. Freddie Mac holds prepayments for an extra
month before passing them through to investors. During that month,
Freddie Mac earns a market rate of return while paying investors at the
mortgage security coupon rate. Float earnings are roughly the same for
both high and low mortgage interest rates, but interest passthrough
payments to investors are much lower on low rate mortgages, increasing
net float income.
(ix) Differences Between Loans on Owner-Occupied and Investor-Owned
Properties
Loans on owner-occupied properties present less credit risk than
loans on investor-owned properties. Simulation 31, presented in Table
16, shows the effects on required capital of adding newly originated,
long-term fixed-rate mortgages that are all investor-owned. Required
capital for loans on investor-owned properties is substantially higher
in all cases because of higher credit losses.
[GRAPHIC] [TIFF OMITTED] TP13AP99.201
2. Commitments
While commitments to purchase mortgages may result in new mortgage
guarantees or new retained mortgages, the risk accepted by the
Enterprise at the time of commitment is comparable to the risk on new
mortgage guarantees. The stress test treats mortgages delivered
pursuant to commitments as guarantees of mortgages that are originated
in the first few months of the stress test at market interest rates.
Hence, no portfolio interest rate risk will be incurred. The mix of
other characteristics of the loans reflects the mix of characteristics
for existing guaranteed loans of the Enterprise that
[[Page 18108]]
were originated during the six months preceding the start of the stress
period.
Simulation 32, shown in Table 17, shows the effects on required
capital of increasing each Enterprise's commitments outstanding in June
1997 by $10 billion. The results are, essentially, an average of the
effects on required capital of a mixture of new loans, in which the
proportions of loans with particular characteristics (including
guarantee fees) match those present in an Enterprise's recently
originated and securitized loans. In the up-rate scenario, the effects
are muted relative to those in the down-rate scenario because the model
assumes that sellers deliver loans for only 75 percent of the
commitment volumes.
[GRAPHIC] [TIFF OMITTED] TP13AP99.202
3. Assets and Liabilities
The Enterprises' other line of business is purchasing mortgages and
mortgage securities for their asset portfolios and funding them with
debt. As holders of mortgages, the Enterprises receive interest income,
incur administrative expenses, and bear the risk of loss if a borrower
defaults. As market interest rates change, the interest rate of a
mortgage becomes more or less favorable, and the value of the mortgage
will change. The Enterprises hedge this risk by issuing callable long-
term debt, which changes in value in a corresponding way. They also
enter into interest rate derivative contracts that further reduce the
overall sensitivity of their income and net worth to interest rate
changes. As a holder of mortgage securities, an Enterprise experiences
cash flows, income, and risks similar to those experienced as a holder
of whole mortgages except that the credit risk is borne by the security
guarantor (usually the Enterprise itself, acting in its other principal
role).
The stress test projects the flows of income and expenses
associated with these assets in much the same way as it does for
mortgage guarantees. However, principal and interest received by an
Enterprise on retained mortgages and mortgage securities is not passed
on to investors, and no credit losses are charged on asset holdings of
mortgage securities guaranteed by either Enterprise or by the
Government National Mortgage Association (Ginnie Mae). In addition, the
stress test projects interest expenses associated with debt and cash
flows associated with derivatives contracts.
a. Assets/Liabilities With Mixed Characteristics Reflecting Enterprise
Portfolios
Table 18 shows the additional capital that would be required in
both scenarios by a general increase in each Enterprise's assets and
liabilities. It is not possible to isolate the average incremental
capital effects of a general increase in an Enterprise's mortgage
assets in the same way that Simulation 1 measured those effects for
guaranteed mortgages. Critical factors in assessing the risk of asset
positions are the characteristics of the debt and equity used to fund
them. However, specific debt and equity issues cannot be matched with
specific assets. It is possible, however, to obtain a measure of the
incremental capital effects of a proportional $10 billion increase in
all of an Enterprise's assets, including non-mortgage assets, and a
simultaneous $10 billion increase in the Enterprise's liabilities and
interest rate derivatives.\45\
---------------------------------------------------------------------------
\45\ The process is indirect, using the results of other
simulations. The increase in required capital for an equal
percentage increase in all of an Enterprise's positions, such that
assets increase by $10 billion, is simply that percentage of the
Enterprise's required capital for the base case simulations for June
1997. This increase includes increases in guarantees and
commitments. The effect of these increases can be removed by
subtracting the incremental effects of the guarantees and
commitments as calculated in Simulations 1 and 32, after making
adjustments for the differences between a $10 billion change in
those factors and a change of the percentage amount used in the
first step.
[GRAPHIC] [TIFF OMITTED] TP13AP99.203
These results reflect some differences between the Enterprises in
asset composition, but, mostly, differences in debt structure and
derivatives use in June 1997. In three of the four cases, the
incremental effects are close to or less
[[Page 18109]]
than the 2.50 percent minimum capital ratio for Enterprise assets. For
both Enterprises, the incremental required capital effects of sold
loans were higher in the down-rate scenario while the effects of asset
holdings and liabilities are higher in the up-rate scenario. Thus, the
combined risks of both types of activities are more balanced with
respect to interest rates than the risks of either type separately.
b. Retained Loans With Specific Identical Risk Characteristics
The simulations discussed below show the effect on required capital
of an increase in mortgage assets that is funded by debt. A first group
of simulations shows how different characteristics of mortgages affect
required capital in each scenario. Five-year, fixed-rate notes were
used to fund mortgage assets in each of these simulations. Different
funding would not have an appreciable effect on the relative results
for mortgages of differing characteristics, as long as the funding was
the same for each. In the second group of simulations, mortgage
characteristics were held constant, while the funding varied among
three alternatives.
The Enterprises have available, and utilize, a much wider range of
funding alternatives than those used in these simulations. These
alternatives include debt (both callable and non-callable) of different
maturities, debt-derivative combinations that create synthetic debt
with various maturity and call characteristics, and debt combined with
swaptions (options on swaps) or with interest rate caps, floors, or
corridors. Other hedging techniques, such as asset swaps, are also
used. The proposed risk-based capital requirements are fully sensitive
to all of these alternatives.
In the Simulations presented in Table 19, $10 billion of retained
unsecuritized loans with specific risk characteristics were added to
each Enterprise's asset portfolio. The assets were funded with $10
billion of five-year notes paying 6.5 percent interest, with no call
options. The mortgages in Simulation 33 have the same characteristics
as those in Simulation 2, except they have not been securitized. They
are newly originated 30-year fixed-rate mortgages, with 80 percent LTV
ratios and 7.5 percent contract interest rates from the West South
Central Census Division. In Simulations 34 through 39, one risk
characteristic (mortgage type, LTV, or age) has been changed from
Simulation 29 to illustrate the relative effects on required capital of
changes in various characteristics.\46\
---------------------------------------------------------------------------
\46\ While these results are for additional retained whole
loans, the effects on required capital of additional holdings of
mortgage security assets, backed by loans with the same
characteristics and funded with the same debt, can be closely
approximated by subtracting the effects of additional guarantees of
loans with those characteristics. (The comparable loan guarantee
simulations are Simulations 2, 17, 20, 12, 13, 5, and 7
respectively.)
[GRAPHIC] [TIFF OMITTED] TP13AP99.204
As the results make clear, using solely five-year fixed-rate debt
to fund mortgages would not be an appropriate funding strategy to guard
against the risk of large, sustained changes in interest rates like
those incorporated in the stress test. When market interest rates
decline, fixed-rate mortgages prepay rapidly, and the five-year debt is
outstanding far longer than most of the mortgages it originally funded.
When market yields rise, fixed-rate mortgages prepay slowly, and the
debt matures long before most of the mortgages are liquidated.
In the up-rate scenario, ARMs with fixed-rate funding reduce
required capital because interest income rises with market yields
(until lifetime caps are reached), while funding costs remain unchanged
during the first five years. Differences in the impact on required
capital of fixed-rate mortgages of different types in the up-rate
scenario primarily reflect differences in credit losses. However, 15-
year loans also benefit from faster amortization, making their loan
lives correspond more closely to the maturity of the debt used to fund
them.
[[Page 18110]]
In the down-rate scenario, ARMs prepay more slowly than FRMs, but
also provide lower interest income. Among fixed-rate types of loans,
four-year-old loans prepay more rapidly than new or eight-year-old
loans. High-LTV loans, on the other hand, prepay slowly because
borrowers lack sufficient equity for refinancing. These differences in
prepayment rates greatly affect the interest rate risk characteristics
of the loans, so that if they are funded with the same liabilities,
four-year old loans with 80 percent LTVs generate higher capital needs
in down-rate scenario than new loans with 95 percent LTVs, despite much
lower credit losses.
The proposed capital requirements are very sensitive to differences
in funding strategies for mortgage assets because of the magnitude of
the interest rate changes in the two scenarios. Table 20 shows the
results of three alternative funding choices for newly originated long-
term FRMs with 80 percent LTVs like those in Simulation 33.
[GRAPHIC] [TIFF OMITTED] TP13AP99.205
Funding long-term FRMs with short-term debt (six-month discount
notes) provides very substantial benefits when interest rates fall. The
debt matures more rapidly than the mortgages, permitting an Enterprise
to continue receiving the original yield on the mortgages, while paying
much lower interest rates. Short-term funding, though, is extremely
costly when interest rates rise because maturing debt must be replaced
at much higher rates. A portfolio of long-term fixed-rate mortgages
funded with short-term debt, such as those held by Fannie Mae and most
thrifts in the late 1970s, would require a capital/asset ratio of well
over 20 percent under the proposed rule.
Funding with long-term debt (ten-year notes with semi-annual
interest payments at 6\3/4\ percent) provides large benefits when
interest rates rise, but is extremely costly when interest rates fall.
Callable long-term debt (ten-year maturity, with a coupon of 7\3/8\
percent, not callable during the first two years) provides benefits in
both scenarios.\47\ The results for different funding mixes can be
approximated by combining the results shown in Table 20 on a weighted
average basis. Thus, for example, in June 1997, the incremental capital
effects of new fixed-rate mortgages funded with 65 percent callable
long-term debt, 19 percent short-term debt, and 16 percent long-term,
non-callable debt would be in a range of 1.2 percent to 2.6 percent for
both Enterprises in both interest rate scenarios. Less callable debt
would be needed to achieve the same result for seasoned loans.
---------------------------------------------------------------------------
\47\ The interest rates of long-term debt used in the
simulations roughly reflect what the average cost of such
instruments would have been in June 1997.
---------------------------------------------------------------------------
4. Administrative Costs
During the stress period, administrative costs depend not only on
the volume of loans held or guaranteed, but also on the rate of
spending in the quarter immediately preceding the start of the stress
period. A higher rate of administrative expense before the stress
period increases costs and depletes capital during the stress period.
In Simulation 43, shown in Table 21, $10 million in annual
administrative expense ($2.5 million at a quarterly rate) was added to
each Enterprise's reported spending in the year preceding the date of
the base case simulations (June 1997).
[GRAPHIC] [TIFF OMITTED] TP13AP99.206
The results in Table 21 show that if Fannie Mae's annual
administrative expense rate had been $1 higher in the year preceding
the stress period, its capital requirement would have been $5.92 higher
in the up-rate scenario and $3.53 higher in the down-rate scenario. The
stress test projects the higher expense rate to continue throughout the
[[Page 18111]]
ten years of the stress period, except that the dollar amount of
additional expense declines in line with the outstanding loan volume.
Thus, in the up-rate scenario, for example, the initial annual $1
increase in the expense rate leads to an additional $7.65 of
administrative expenses during the stress period. Discounting, taxes,
and dividends reduce the incremental required capital to $5.92, even
after the 30 percent management and operations risk supplement.
Required capital increases more in the up-rate scenario than the down-
rate scenario because administrative expense is tied in the stress test
to outstanding loan volumes, which are larger in the up-rate scenario.
The effect of increased administrative expenses on required capital
is lower for Freddie Mac in both interest rate scenarios. This is true
partly because Freddie Mac's mortgages have slightly shorter lives in
both interest rate scenarios, but more importantly because Fannie Mae
has disproportionately larger commitments outstanding at the start of
the stress period. As commitments are transformed into loans during the
early months of the stress period, Fannie Mae's overall loan balances
rise relative to initial balances by more than Freddie Mac's. This
effect is less significant in the up-rate scenario because only 75
percent of commitments become loans. However, Freddie Mac's costs in
the up-rate scenario are reduced by taxes throughout the stress period,
while Fannie Mae's are not. Therefore, Freddie Mac's administrative
expense rate has a smaller effect on required capital in both interest
rate scenarios.
5. External Economic Conditions
a. House Prices
Stress test results are also greatly affected by changes in
external economic conditions. Seasoned mortgages in the base case
simulations for June 1997 benefited from modest, but steady average
house price appreciation of about three percent per year during the
time between origination and the beginning of the stress period. In
Simulations 46 and 47, shown in Table 22, the house price index was
reduced by one percent and five percent, respectively, in the quarter
immediately preceding the stress period (1997 Q2). That is, house price
appreciation rates between the first and second quarters of 1997 were
assumed to be one percentage point or five percentage points (4 or 20
percentage points at an annual rate) less than they actually were.
Subsequent house price appreciation rates are the same as in previous
simulations.
[GRAPHIC] [TIFF OMITTED] TP13AP99.207
When house prices are decreased by one percent, credit losses for
each Enterprise increase by four to five percent in the up-rate
scenario and by about seven percent in the down-rate scenario. The
increases in credit losses when house prices are decreased by five
percent are about five times as large as they are for a one percent
house price decrease. The increases in incremental capital in both
simulations are larger in the down-rate scenario because the decrease
in house prices slows prepayment rates in that scenario, owing to
higher probabilities of negative equity. Slower prepayment rates
increase the volume of mortgages exposed to the risk of default. While
loans also prepay more slowly in the up-rate scenario, prepayment rates
in the base case simulation for that scenario are already so slow that
a similar percentage change has little absolute effect.
The slowing of prepayment rates with lower house prices in the
down-rate scenario also produces two benefits that offset much of the
increase in loan losses: guarantee fee income and net interest income
increase. The key factor causing the effects on required capital to be
larger in the down-rate scenario is that discount rates are lower in
that scenario, so the present value of similar additional credit losses
is greater.
Differences in the changes in required capital between the
Enterprises primarily reflect lower additional credit losses for
Freddie Mac. Fannie Mae's losses are higher because its owned or
guaranteed loan volume was about 45 percent larger than Freddie Mac's
in June 1997 and its credit losses per dollar of loans are 11 to 14
percent higher in the simulations, owing to a somewhat riskier mix of
loans.
b. Market Interest Rates
The behavior of interest rates in the months before the starting
date of the stress test can also have a significant effect on required
capital. In the simulations shown in Table 23, all market yields were
assumed to be 200 basis points higher (Simulation 46), or lower
(Simulation 47) in the month preceding the stress test period (June
1997) than they actually were.\48\ The principal means by which this
change in market yields affects required capital is through the change
it causes in market interest rates during the last nine years of the
stress test.\49\
---------------------------------------------------------------------------
\48\ No changes were made to interest rates on asset, liability,
or off-balance sheet positions that had been put in place during the
month, but they constitute a small share of total positions, and the
effects of adjusting interest rates for those positions would have
been largely offsetting. Nor were any changes made to Enterprise
hedge positions that they might have made had market yields actually
changed.
\49\ In the circumstances of June 1997 (or any other time since
September 1991), the applicable statutory rule for determining the
change in the ten-year constant maturity Treasury yield during the
stress period is that it increases by 75 percent or decreases by 50
percent from the average over the preceding nine months. If interest
rates were 200 basis points higher in June 1997, stress test rates
would have risen to a level 200 9 x 1.75 = 39 basis
points higher for the last nine years in the up-rate scenario. And,
in the down-rate scenario, rates would have decreased to a level 200
9 x 0.50 = 11 basis points higher. Similarly, if interest
rates were 200 basis points lower in June 1997, stress test rates
would have been 39 basis points lower in the last nine years of the
up-rate scenario and would have fallen to a level 11 basis points
lower in the last nine years of the down-rate scenario. These
differences are incorporated in Simulations 46 and 47.
[[Page 18112]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.208
In Simulation 46, the hypothetical increases in June 1997 yields
make the stress test more severe in the up-rate scenario and less
severe in the down-rate scenario. Simulation 47 does the reverse. The
size of the effects is much greater for Fannie Mae because its asset
size was roughly double Freddie Mac's at the time, and because Fannie
Mae's interest rate risk was less fully hedged then Freddie Mac's.
Although changes in net interest income accounted for nearly all of the
change in required capital, differences in prepayment rates in the
down-rate scenarios of both simulations affected required capital
through changes in other income and expense categories. Lower
prepayment rates in Simulation 46 increased credit losses, but also
increased guarantee fees. Higher prepayment rates in Simulation 47
decreased credit losses and guarantee fees.
c. Sensitivity to Risk Characteristics in Different Economic
Environments
The results of the sensitivity analysis discussed above are
dependent on the risk structure of the Enterprises and the economic
conditions of June 1997. For example, as discussed above, credit losses
on seasoned loans vary depending on house price behavior between the
time of origination and the start of the stress test. At higher
interest rate levels, the consequences of imperfectly matched assets
and liabilities would be greater because stress test changes in
interest rates would be larger. At lower interest rate levels, the
effects would be smaller. Different Enterprise hedging strategies could
affect reported sensitivities because they could result in a different
pattern of profits and losses during the stress period, which could
affect the role of taxes. Changes in common stock dividend payouts
could affect the impact of dividends during the first year of the
simulations.
C. Implications of the Proposed Rule
The Enterprises perform an important role in the nation's housing
finance system. Although the current risk of an Enterprise failure is
small, the continued financial health of the Enterprises cannot be
taken for granted. Over the past two decades, failures of financial
institutions have been commonplace, including more than 2900 banks and
thrifts and a number of securities firms. The risks associated with
Fannie Mae and Freddie Mac differ in some important ways from those
associated with banks, thrifts, and securities firms. However,
government sponsored enterprises are not immune to failure. Fannie Mae
encountered serious financial difficulty in the early 1980s, recovering
in large part because of a fortuitous decline in interest rates, and
the Farm Credit System experienced serious problems later in the
decade. Because of the Enterprises' key role and important public
mission, Congress created OFHEO to ensure their safe and sound
operation. The current combined obligations of the Enterprises amount
to more than $1.7 trillion, and unlike banks, thrifts, and securities
firms, no Enterprise obligations are backed by an insurance fund that
could contribute toward meeting creditor claims.
The risk-based capital rule (in conjunction with OFHEO's other
regulatory tools) is intended to reduce the risk of financial failure
of an Enterprise. The rule can contribute to that goal by requiring the
Enterprises to hold more capital or take less risk than they otherwise
would in some or most potential circumstances, particularly those
circumstances in which the danger of failure is greatest. In
circumstances in which some capital or risk adjustment is necessary,
the rule gives an Enterprise the flexibility to choose whether more
capital, less risk, or a combination of the two best suits its business
needs.
OFHEO believes that the proposed rule would effectively serve its
intended role. By promoting the Enterprises' safety and soundness, the
regulation promotes their ability to continue to carry out their public
purposes.\50\ These include providing stability in the secondary market
for residential mortgages and providing access to mortgage credit in
central cities, rural areas, and underserved areas.
---------------------------------------------------------------------------
\50\ 1992 Act, section 1302(2) (12 U.S.C. 4501(2)).
---------------------------------------------------------------------------
Capital reduces the risk and costs of failure by absorbing losses.
For most firms, debt markets provide strong capital discipline,
penalizing a firm that is excessively leveraged with higher borrowing
costs. That discipline is largely lacking for the Enterprises because
of their government sponsored enterprise status. The lack of normal
market discipline makes capital requirements particularly important for
the Enterprises.
The minimum capital regulation, currently in place for the
Enterprises, provides important protection against failure. It requires
the Enterprises to have a minimally acceptable level of capital in
relation to their overall size, regardless of their measurable risk.
The establishment of the minimum capital standard was accompanied by
considerable increases in capital at both Enterprises. Because,
however, it is based on simple leverage ratios, it will not be
sufficient if an Enterprise chooses to take risky financial positions
or if market conditions move adversely and increase the risk of what
had been less risky positions. By contrast, the proposed rule is quite
sensitive to risk. It would require an Enterprise to increase capital
when risk rises, well before the potential adverse
[[Page 18113]]
consequences of the rise would be reflected in the Enterprise's
financial statements. Each of the two capital rules is an essential
complement to the other.
1. Capital Requirements Under the Proposed Rule
Consistent with the purpose of reducing the risk of Enterprise
failure, the proposed rule can be expected to influence how the
Enterprises manage their risk and the amount of capital they hold.
Table 24 shows actual total capital (amounts available to meet the
risk-based capital requirement) and required total capital under the
proposed rule for two dates: September 30, 1996 and June 30, 1997.\51\
It also shows actual core capital (amounts available to meet the
minimum capital requirement) and required core capital on the same
dates. The difference between total capital and core capital is that
total capital includes general loss reserves, while core capital does
not.
---------------------------------------------------------------------------
\51\ These results include estimated effects on required total
capital for three provisions of the proposed rule that require
credit ratings: credit losses on non-mortgage investments; on
derivative contracts; and on rated mortgage-related securities, such
as mortgage revenue bonds. OFHEO assumed that 50 percent of non-
mortgage investments are rated AAA, 35 percent are rated AA, and 15
percent are rated A. The percentages for derivative contracts are
85, 15, and 0, respectively; and those for rated mortgage-related
securities are 70, 30, and 0, respectively. The results do not
reflect the effects of master netting agreements, nor haircuts on
foreign-denominated contracts. Multifamily credit enhancements,
other than those for Fannie Mae's DUS product are not modeled
explicitly, but are assumed to reduce loss severities by 15.9
percentage points.
[GRAPHIC] [TIFF OMITTED] TP13AP99.209
Table 25 shows the surplus or deficit of total capital for both
interest rate scenarios. The risk-based capital requirement for an
Enterprise is based on the scenario that would result in the greatest
deficit or smallest surplus. To meet the requirement, an Enterprise
must not have a capital deficit in either scenario. Freddie Mac would
have had a risk-based capital surplus of 28 percent on the 1996 date
and 19 percent in 1997, while Fannie Mae would have had a deficit on
each date of 21 percent. In contrast, both firms met the existing
minimum capital standard on both dates, with surpluses ranging from 4
percent to 11 percent. Thus, the risk-based capital requirement would
have been much higher than the minimum capital requirement for Fannie
Mae, even after taking account of the differences in the definition of
capital under the two standards. For Freddie Mac, however, the minimum
capital requirement would have been higher than the risk-based capital
requirement. Thus, the risk-based standard would not have imposed any
additional requirement on Freddie Mac on those dates. The primary
reason Fannie Mae's risk-based capital requirement would have exceeded
its minimum capital requirement, while Freddie Mac's would not, is that
Freddie Mac's asset/liability structure was more fully hedged against
interest rate risk than Fannie Mae's.
[[Page 18114]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.210
Risk-based capital requirements in the future may vary
significantly, depending not only on the Enterprises' assets and
obligations, but also on contemporary economic conditions. Declines in
house prices in the years preceding the starting date of the stress
test can greatly raise capital requirements under the proposed rule,
and rapid house price appreciation during these years can greatly
reduce them. Unhedged interest rate exposures would require greater
capital when interest rates are higher at the start of the stress
period because changes in interest rates during the stress period will
be greater. The reverse is true when interest rates are lower. Economic
environments entailing greater than usual uncertainty about future
interest rates or mortgage defaults will be accompanied by higher costs
for hedges, such as callable debt or credit enhancements. In the
absence of a risk-based capital standard, an Enterprise might choose to
maintain capital and hedges that would be sufficient to meet the
proposed standard in low risk environments, but might not do so in high
risk environments owing to the higher cost of capital and hedges in
such environments.
2. Enterprise Adjustments To Meet the Proposed Standard
An Enterprise with capital and risk preferences that are not
consistent with the proposed standard could adjust to the standard by
either increasing capital or decreasing risk or both. Capital can be
increased by reducing share repurchases, adjusting dividends, or
issuing new equity shares. Enterprise risk can be reduced by increasing
the use of interest rate and credit risk hedges, after risk is taken
on, or by reducing the amount of risk taken on.
Financial markets currently provide a wide range of hedges against
interest rate risk. These include, among others: callable long-term
debt, caps and floors, and swaps and swaptions. Adding interest rate
risk hedges may frequently be cheaper than increasing equity. For
example, based on the differences in results of Simulations 40, 41, and
42 shown in Table 20, Fannie Mae could have met the proposed standard
in June 1997 by issuing $22 billion of callable ten-year notes and
using the proceeds to pay off $14 billion of short-term debt and
repurchase $8 billion of ten-year notes.\52\ Given the market yields at
that time, such a change in debt structure would have cost less than
$200 million on an annual basis, after taxes. However, because this
debt restructuring would have provided substantial benefits in terms of
reduced risk, the net cost would have been much lower.
---------------------------------------------------------------------------
\52\ The interest rates of long-term debt used in the
simulations roughly reflect what the average cost of such
instruments would have been in June 1997.
---------------------------------------------------------------------------
Changes in an Enterprise's asset/liability structure to reduce
interest rate risk, such as the one described in the above example, may
be much cheaper than raising new equity. If the annual cost of equity
capital is assumed to be 15 percent, the net cost of raising sufficient
equity would have been roughly $385 million.\53\ Other forms of
liability restructuring, or changes in the interest rate risk
characteristics of the assets, might have resulted in lower costs than
those estimated here for hypothetical changes in debt structure. Fannie
Mae anticipated the likelihood of such opportunities in its comment on
OFHEO's ANPR: ``* * * if the [mortgage] portfolio is in a position
where its risk-based capital requirement exceeds its actual capital,
the practical remedy would be to change the portfolio's asset/liability
structure so that this is no longer the case.'' An alternative way for
an Enterprise to reduce its interest rate risk is simply to reduce the
size of its asset portfolio. Given the high profitability of those
portfolios in recent years, that currently would not be a likely
choice.
---------------------------------------------------------------------------
\53\ In its analysis supporting its affordable housing goal
rule, HUD used an estimate for the cost of equity capital of 17
percent, but subsequent increases in price-earnings ratios suggest a
smaller number for more recent dates. The cost calculation assumes
that the additional equity would have replaced an equal amount of
debt.
---------------------------------------------------------------------------
Increasingly, credit risk can also be hedged in financial markets.
Freddie Mac's 1998 MODERNS transaction effectively transferred a
portion of the credit risk on its 1996 mortgage purchases to investors
in the new securities.\54\ Further development of the credit
derivatives market may provide additional opportunities for
transferring credit risk in the future. An Enterprise can also reduce
its credit risk by requiring or acquiring more credit enhancements. As
an example, the Enterprises increased requirements for mortgage
insurance on 95 percent LTV loans starting in 1995.
---------------------------------------------------------------------------
\54\ Investor returns on the securities are dependent on the
rate of defaults in a pool of mortgages representing 17.4 percent of
Freddie Mac's single family, 30-year FRMs purchased in 1996.
---------------------------------------------------------------------------
Finally, an Enterprise could adjust to a capital shortage by
curtailing the size of its mortgage guarantee business. Such a measure
is likely to be taken only as a last resort, as that business is the
primary means by which an Enterprise fulfills its fundamental public
purposes. As long as that business is profitable, an Enterprise is
likely to prefer to restructure its asset/liability positions, obtain
more credit risk hedges, or, if necessary, raise additional capital. If
the Enterprise is financially safe and sound, raising additional equity
capital should not be difficult. Because the proposed rule should help
ensure the Enterprise's continued healthy financial condition, the rule
would make it less, rather than more, likely that the Enterprise will
need to restrict its activities.
3. Guarantee Fees
It is unlikely that the proposed rule will have any material
effects on the general level of guarantee fees charged by the
Enterprises. The stress test results make it particularly unlikely that
the rule would have any effects on guarantee fees in economic
environments like those of the recent
[[Page 18115]]
past. Freddie Mac would have met the risk-based standard in 1996 and
1997 by substantial margins, without any changes to its balance sheet
or business operations. Thus, the risk-based capital standard would not
have given Freddie Mac any cause to raise guarantee fee levels. Fannie
Mae would not have been able to, if it wished to maintain its
competitive position. In the future, there may be circumstances in
which the capital or risk positions of both Enterprises are affected
simultaneously by the risk-based standard. The analysis of such cases
is more complicated. However, the duopolistic structure of the
secondary mortgage market and the generally small impact of the
guarantee business on required capital make it unlikely that the
standard would affect guarantee fees in those circumstances, either.
Guarantee fees compensate the Enterprises for assuming credit risk
on the mortgages they purchase in the secondary market. They may be
explicit, as they are for securitized loans, or implicit, as they are
for loans purchased for Enterprise portfolios. These fees primarily
cover expected credit losses and operating expenses, but include a
return to the capital needed to protect against more severe credit
losses in adverse environments. The need to provide such a return
effectively makes capital a component of cost in the Enterprises'
secondary market activities.
In a fully competitive market, a regulation (such as a capital
regulation) that raises the marginal costs of all firms in that market
would result in higher prices (guarantee fees in this case). However,
the secondary mortgage market is not fully competitive.\55\ Fannie Mae
and Freddie Mac constitute virtually the entire buy side of the
secondary market for fixed-rate conforming, conventional mortgages,
making that market a duopoly.\56\ In a duopoly, the two firms generally
exercise market power by charging prices (the guarantee fee) in excess
of marginal cost, and thereby recognizing economic profits.
---------------------------------------------------------------------------
\55\ For a fuller discussion of secondary mortgage market
structure and behavior, see Benjamin E. Hermalin and Dwight M.
Jaffe, ``The Privatization of Fannie Mae and Freddie Mac:
Implications for Mortgage Industry Structure,'' in Studies on
Privatizing Fannie Mae and Freddie Mac, U.S. Department of Housing
and Urban Development, May 1996. This paper was jointly commissioned
by HUD, the Department of the Treasury, the General Accounting
Office, and the Congressional Budget Office.
\56\ The ``buy side'' terminology here is traditional but
confusing. The Enterprises are either buying mortgages or selling
guarantees. Either way, they are charging implicit or explicit fees
for assuming credit risk.
---------------------------------------------------------------------------
In theory, the guarantee fee charged by Fannie Mae and Freddie Mac
may range between the perfectly competitive rate (where the fee equals
the firms' marginal cost) and the monopoly rate (where the fee
maximizes the two firms' joint profits as if they were operating as a
cartel). If the fee at which other firms may enter the market is less
than the monopoly fee, then the maximum fee would be that at which
entry would take place.
The Enterprises' current guarantee fees reflect the profit-
maximizing decisions of both Enterprises. These decisions are affected
by the degree of competition between the two firms, the threat of entry
by other firms, and activities necessary to maintain or enhance the
value of their public charters. The current level of guarantee fees
already reflects the maximum guarantee fees that each Enterprise feels
it can charge without reducing long-run profits. If this were not the
case, Enterprise shareholders likely would object. In such
circumstances, a small increase in capital (or any other) cost is
unlikely to affect guarantee fees. Only if the cost increase was
sufficiently large to raise marginal cost (including an adequate return
to attract capital) above the current fee level, would a fee increase
reasonably be expected.
The Treasury Department and the Congressional Budget Office
estimated in 1996 that the Enterprises collected roughly five basis
points (0.05 percent) in fees for their mortgage-backed security
guarantees above what they would need to recover costs plus a normal
profit margin.\57\ After taxes (at an effective rate of 30 percent),
that amounts to 3.5 basis points. A risk-based capital standard that
raised the capital costs associated with the Enterprises' guarantee
business by less than that amount would still allow the Enterprises to
earn returns above a normal profit margin.
---------------------------------------------------------------------------
\57\ U.S. Department of the Treasury, The Government Sponsorship
of the Federal National Mortgage Association and the Federal Home
Loan Mortgage Corporation, July 11, 1996; The Congressional Budget
Office, Assessing the Public Costs and Benefits of Fannie Mae and
Freddie Mac, May, 1996.
---------------------------------------------------------------------------
If a new capital standard required an Enterprise to increase its
equity when it increased its guarantee business, its capital cost per
dollar of new guarantee business would be the amount of additional
capital required times the cost of new equity capital, perhaps 15
percent. The proposed rule, however, provides an alternative to raising
equity, which is to reduce some other risk. As shown in the previous
section, Fannie Mae could meet an overall higher capital requirement of
$3.68 billion at an after-tax cost of less than $200 million in June
1997. The cost per dollar of additional capital requirements was only
about 5.4 cents (0.20 3.68). An additional dollar of capital
requirements associated with new guarantee business could be met in the
same way. Based on that cost of capital, if an additional dollar of
guarantee business caused required capital under the new standard to be
65 basis points greater than under the existing standard, the
additional capital cost would be only as great as the duopoly surplus
margin of 3.5 basis points (65 x .054 = 3.5).
In the absence of a risk-based capital standard, regulatory capital
costs are based on the existing minimum capital leverage ratio for
mortgage-backed security guarantees, which is 0.45 percent (45 basis
points). A comparison with the incremental capital required for sold
loans under the risk-based capital requirement must take into account
that the leverage requirement can be met only with equity (core)
capital, while the risk-based requirement can be met with both equity
and reserves (total capital). Reserves for losses on mortgage-backed
security guarantees average about seven basis points per dollar of
guarantees at both Enterprises, so the comparable minimum capital
requirement in terms of total capital is 52 basis points. Thus, a risk-
based capital standard could potentially raise the incremental amount
of total capital required for sold loans to as much as 117 basis points
(52 + 65) and still allow the Enterprises to earn sufficient profits to
continue to attract capital.
Even greater increases would be unlikely to affect guarantee fees
in circumstances when the capital and risk decisions of one or both
Enterprises are unaffected by the risk-based standard, as was
presumably the case for Freddie Mac on the two recent dates for which
risk-based capital calculations have been performed. If the risk-based
standard were binding (affected capital or risk decisions) for only one
of the Enterprises, then, even if its incremental risk-based
requirements for sold loans were very much higher than the minimum
capital ratio, it would be difficult for that Enterprise to raise
guarantee fees independently. Doing so likely would cause it to lose
market share and profits to the other Enterprise.
Even if the risk-based standard were binding on both Enterprises,
it appears unlikely that the proposed standard would raise the capital
required for the Enterprises' mortgage guarantee business to as much as
117 basis points. The results of a simulated increase in
[[Page 18116]]
overall MBS guarantee volumes, shown in Table 6, indicate that the
incremental capital required in 1997 for the up-rate scenario of the
risk-based standard was well below the 52 basis points needed to meet
the minimum capital standard. In the down-rate scenario, incremental
capital of as much as 89 basis points would have been needed, but that
is still substantially below the 117 basis points level that
potentially would trigger a rise in guarantee fees.
While the results referred to in Table 5 are informative, an
Enterprise evaluating the capital costs associated with its mortgage
guarantee business would properly focus on its prospective costs at
future dates. To do so, it would want to estimate the likelihood of its
being bound by the risk-based standard in the future, and if it thought
it would be bound, the relative likelihood of being bound by the up-
rate and down-rate scenarios. It would also want to make informed
guesses about the other Enterprise's estimations on its own behalf.
Finally, it would want to estimate the likelihood of significantly
higher incremental capital requirements for sold loans under the risk-
based standard.
These incremental requirements will be affected by the pace of
house price appreciation in the years preceding the date of capital
calculation. The figures in Table 5 reflect annual appreciation of
about three percent, lower than long-run historical averages. If an
Enterprise anticipated stagnant or declining house prices over an
extended period of time, and if it believed both itself and the other
Enterprise likely would be bound by the risk-based standard,
particularly the down-rate scenario, it might have an incentive to
raise guarantee fees. In such a circumstance, its expected losses would
also rise, and likely by far more than its capital costs. The higher
expected losses would, in that case, be the principal cause of higher
fees.
A riskier interest rate environment could also affect projected
capital costs. If the cost of interest rate risk hedges rose
dramatically, so that it became cheaper to meet shortfalls in required
capital by raising new equity than by increasing interest rate hedges,
any increase in capital required by an Enterprise's sold loans would be
more costly and more likely to lead to a small increase in guarantee
fees. However, providing adequate protection in unusually risky
economic environments, such as those with much higher interest rate
hedging costs or persistent weakness of house prices is a fundamental
purpose of the risk-based capital standard.
OFHEO has also considered the possibility that the proposed
standard, while not affecting the general level of guarantee fees,
could affect the fees charged directly or indirectly on loans made to
low income borrowers. Such effects are unlikely and would, in any
event, be minimal. Consequently, the risk-based capital standard will
not significantly affect the Enterprises' ability to purchase
affordable housing loans. These conclusions are based on several
considerations. First, the capital surpluses that Freddie Mac would
have held in 1996 and 1997 under the rule show that no changes in any
Enterprise fees or loan-purchase practices would have been justified in
recent economic environments.
Second, with respect to potentially more adverse environments, the
capital cost of single family loans meeting the Enterprises' affordable
housing goals should not be materially different, on average, from the
cost of other loans. The stress test makes no specific distinctions
among loans to different income groups. However, the stress test does
distinguish single family loans according to LTV class and some
Enterprise affordable products are high LTV loans. The simulation
results in Section II. B., Sensitivity of Capital Requirements to Risk,
show that high LTV single family loans are generally riskier and affect
risk-based capital requirements more than other loans. However, the
overall LTV distribution of single family loans purchased by Fannie Mae
and Freddie Mac for low-and moderate-income borrowers (borrowers with
less than area median income) is practically the same as the LTV
distribution of all their purchased loans. In fact, only a small
percentage of the loans to low- and moderate-income borrowers purchased
by the Enterprises are high LTV loans (those with LTV ratios above 90
percent).
Third, while high LTV loans have much higher than average risk, the
simulation results overstate the capital implications of those loans.
The results of Simulations 13 and 15, in Table 12, show incremental
capital required under the risk-based standard for new and four-year-
old loans, as of June 1997. For a weighted average of Enterprise loans
guaranteed at that time, these incremental requirements were about 170
basis points above the comparable minimum capital ratio in the up-rate
scenario, and about 325 basis points above in the down-rate scenario.
Those differences in capital required, however, overstate the impact of
high LTV loans because they assume only an average level of guarantee
fees. As discussed earlier, the Enterprises generally charge higher
fees implicitly on such loans by adjusting the average fees charged to
lenders according to the average risk of the loans they deliver. And as
shown by the comparison of Simulations 2 and 3, in Table 8, differences
in guarantee fees affect incremental capital requirements. The
overstatement may be increased by the assumption that the Enterprises
have priced these loans based on the incremental capital needed to meet
the minimum standard. Both Enterprises use internal capital models that
reflect the higher risk of high LTV loans and already may incorporate
higher capital costs into the implicit fees charged for these loans.
Fourth, the capital implications of multifamily loans, which
predominately benefit low- and moderate-income households, are mixed
and serve, in some circumstances, as hedges for other high-risk loans.
Simulations 22 to 25 show a wide variety of incremental capital
requirements under the risk-based standard for June 1997. On a weighted
average basis, accepting credit risk on multifamily loans lowered risk-
based requirements in the down-rate scenario and raised them somewhat
more than minimum capital requirements in the up-rate scenario. The
results in the down-rate scenario are the reverse of the pattern for
high LTV single family loans, so that higher costs on high LTV single
family loans are substantially offset by lower costs on multifamily
loans. In the up-rate scenario, the potential effects of high LTV loans
and multifamily loans are similar, but not large.
Finally, even if the proposed rule did require some additional
capital against a portion of the Enterprises affordable housing
activities, such a requirement would be consistent with the
Enterprises' charters and public mission. The Enterprises' charters
specifically state that the return on required lending to low-and
moderate-income borrowers may be less than the return earned on other
activities.
4. Mortgage Interest Rates
The primary effects of the Enterprises' activities on mortgage
interest rates occur through their roles as mortgage security
guarantors. Mortgage security yields are determined in capital markets,
and the interest rates borrowers pay reflect those yields plus the
margins retained by the Enterprises, as guarantee fees, and those
retained by lenders and servicers. Because of the dominant role of the
Enterprises in the market for conforming, single family mortgages,
increases in their guarantee fees would raise lenders' costs and
translate fairly directly to changes in borrowers' costs.
[[Page 18117]]
However, because the proposed rule likely will have no material effect
on guarantee fees, it would not have a significant effect on mortgage
rates through the Enterprises' roles as mortgage guarantors.
As investors in mortgages and mortgage securities, the Enterprises
may also affect mortgage rates indirectly. They now hold roughly an
eighth of all conforming, single family mortgages, and massive changes
in their purchase volumes could have some effect, at least temporarily,
on prices in that market. However, the Enterprises do not dominate the
mortgage investment asset market in the same way that they dominate the
market for guarantees on conforming loans. Consequently, the effects on
mortgage security yields of even substantial changes in their
investment in mortgage securities would be small. Furthermore, the
proposed rule is unlikely to have a substantial effect on Enterprises'
purchases of mortgage assets. Freddie Mac added roughly $100 billion to
its portfolio in the four years preceding the June 1997 simulations and
still easily met the requirements of the proposed rule. Thus, it is
unlikely that the proposed rule would affect the mortgage interest
rates paid by borrowers through the Enterprises' roles as mortgage
investors, either.
III. Issues, Alternatives Considered
A. Mortgage Performance
The 1992 Act requires the risk-based capital test to subject the
Enterprises to specified adverse credit and interest rate risk
conditions to determine the level of capital needed to survive a
hypothetical ten-year stress period. The 1992 Act does not specifically
refer to mortgage performance, but rather discusses the credit-risk
portion of the stress test as including rates of mortgage default and
loss severity. As a convenience, OFHEO used the term ``mortgage
performance'' in the ANPR to facilitate discussion of the essential
elements of credit risk, mortgage default and loss severity, as well as
mortgage prepayment, a key element of interest rate risk. The 1992
Act's requirement to determine a prepayment experience consistent with
the stress period is also relevant to credit risk, because loans that
are paid off prior to maturity affect default rates by reducing the
number of loans that have the potential to default and by increasing
the proportion of loans likely to default. Together, default,
prepayment, and loss severity define how a portfolio of mortgages will
perform in the proposed stress test. That performance is a key element
in determining the ability of an Enterprise to withstand the economic
shocks imposed by the stress test.
To determine the level of capital needed to survive the stress
test, the proposed regulation uses a monthly cash flow model to project
the performance of each Enterprise during the stress period. Underlying
the simulation of mortgage and mortgage security cash flows are models
that project mortgage performance during the stress period.
This section discusses the issues, alternative approaches and
related ANPR comments that were considered by OFHEO in developing
models to project mortgage performance under economic conditions
specified in the l992 Act. Section III. A. 1., Statutory Requirements
describes relevant statutory requirements. Section III. A. 2., Overview
of Mortgage Performance, explains how mortgage performance is measured
and projected in the stress test. Next, in section III. A. 3.,
Statistical Models of Mortgage Performance, through section III. A. 7.,
Relating Losses to the Benchmark Loss Experience, the issues
encountered by OFHEO in developing models of mortgage performance,
along with relevant comments received in response to the ANPR, are
discussed. Section III. A. 3., Statistical Models of Mortgage
Performance, discusses OFHEO's decision to employ statistical models to
predict default, prepayment, and severity rates. Section III. A. 4.,
General Methodological Issues, reviews general methodological issues
encountered in making product distinctions and developing loan and
property value data for use in estimating the statistical models and in
applying those models in the stress test. Section III. A. 5., Default/
Prepayment Issues, details the construction of the default and
prepayment models, including use of conditional rates of default and
prepayment, use of joint models of default and prepayment, and choice
of the explanatory variables used in the models. Section III. A. 6.,
Loss Severity, moves from default and prepayment to issues encountered
in modeling loss severity rates. Section III. A. 7., Relating Losses to
the Benchmark Loss Experience, discusses issues arising from the
statutory direction to reasonably relate stress test losses to the
benchmark loss experience.
1. Statutory Requirements
The 1992 Act mandates a stress test based on a regional recession
involving the highest rates of default and loss severity experienced
during a period of at least two years in an area containing at least
five percent of the total U.S. population.\58\ This mandate required
identifying a benchmark loss experience, which is the default and
severity behavior of mortgage loans, in a place and time meeting
statutory requirements, that resulted in the highest loss rate for any
such place and time.\59\ In this context, default and severity behavior
means the frequency, timing, and magnitude of losses on mortgage loans,
given the specific characteristics of those loans and the economic
circumstances affecting those losses. The 1992 Act requires that
default and severity rates in the stress test be reasonably related to
this benchmark loss experience. In contrast, the 1992 Act does not
prescribe any particular experience for the third key component of
mortgage performance, prepayment. Rather, the Act requires that the
Director determine prepayment levels, ``on the basis of available
information, to be most consistent with the stress period.'' \60\
---------------------------------------------------------------------------
\58\ 1992 Act, section 1361(a)(1) (12 U.S.C. 4611(a)(1)).
\59\ See 61 FR 29592, June 11, 1996, in which OFHEO proposed
procedures for establishing the benchmark loss experience.
\60\ 1992 Act, section 1361(b)(2) (12 U.S.C. 4611(b)(2)).
---------------------------------------------------------------------------
The 1992 Act requires the Director to take into account appropriate
distinctions among mortgage product types and differences in loan
seasoning. It also authorizes the Director to also take into account
any other factors that the Director deems appropriate.\61\ The statute
defines the term ``seasoning'' as ``the change over time in the ratio
of the unpaid principal balance of a mortgage to the value of the
property by which such mortgage loan is secured.'' \62\ The importance
of seasoning is that a decline in a property's value can result in
negative equity, the factor most predictive of rates of default.
---------------------------------------------------------------------------
\61\ 1992 Act, section 1361(b)(1) (12 U.S.C. 4611(b)(1)).
\62\ 1992 Act, section 1361(d)(1) (12 U.S.C. 4611(d)(1)).
---------------------------------------------------------------------------
The 1992 Act defines mortgage product type as a classification of
one or more mortgage products having similar characteristics with
respect to the property securing the loan, the interest rate, the
priority of the lien, the term of the mortgage, the owner of the
property (owner-occupant vs. investor), the nature of the amortization
schedule, and any other characteristics as the Director may determine.
Specifically, the 1992 Act requires OFHEO to take into account
distinctions between different mortgage types, such as: (1) properties
consisting of 1-4 residential units and those containing more than four
units;
[[Page 18118]]
(2) fixed and adjustable interest rates; (3) first and second liens;
(4) terms of 1-15 years, terms of 16-30 years and terms of more than 30
years; (5) owner occupants and investors; and (6) fully amortizing
loans and loans that are not fully amortizing.
The 1992 Act prescribes two interest rate scenarios, one with rates
falling and the other with rates rising.\63\ In each scenario, the ten-
year constant maturity Treasury yield (CMT) experiences a significant
change during the first year of the stress test, and then remains at
the new level during the remaining nine years of the stress test. The
capital requirement for each Enterprise is based on the scenario with
the more adverse impact.\64\ The 1992 Act recognizes that interest
rates are related to credit risk as well as interest rate risk,
specifically requiring that credit losses be adjusted for a
correspondingly higher rate of general price inflation if applying the
stress test results in an increase of more than 50 percent in the ten-
year CMT.\65\
---------------------------------------------------------------------------
\63\ 1992 Act, section 1361(a)(2) (12 U.S.C. 4611(a)(2)).
\64\ 1992 Act, section 1361(a)(2) (12 U.S.C. 4611(a)(2)).
\65\ 1992 Act, section 1361(a)(2)(E) (12 U.S.C. 4611(a)(2)(E)).
---------------------------------------------------------------------------
2. Overview of Mortgage Performance
The amount of capital needed to survive the stress conditions
prescribed by statute is determined by the overall financial
performance of the Enterprises' starting books of business, including
all assets, liabilities, and off-balance sheet obligations, under the
stress conditions. Mortgage performance contributes to the overall
financial performance of an Enterprise during the stress period,
because various sources of income and expense reflected on an
Enterprise's income statement depend directly on mortgage performance.
For example, guarantee fee income on securitized loans, net interest
income on retained loans and securities, and losses on defaulting loans
(offset by the receipt of private mortgage insurance payments and other
third-party credit enhancements) all depend on the projected default
and prepayment behavior of the underlying mortgage assets.
For purposes of the proposed regulation, mortgage performance is a
function of the survival or termination of loans and, ultimately, the
associated cash flows. Loan terminations can occur either through
default (borrower failure to pay) or through prepayment (early payment
in full). Prepayments have a significant impact on credit risk, because
they affect the timing and rates of default. Prepayments also affect
Enterprise income, because they cut off the income stream from interest
payments or guarantee fees. Defaults likewise cut off the income
stream, and, in addition, result directly in credit losses.
To understand how the stress test generates and uses mortgage
performance information, the test may be viewed as comprised of three
elements--models, stress test specifications, and data inputs. In the
context of mortgage performance, the models are sets of equations
designed to predict the performance of any group of Enterprise
mortgages under any given set of economic circumstances. The model
equations themselves are ``estimated'' based upon OFHEO's historical
database of mortgage information to predict the most likely default and
severity rates for any given group of mortgages under any given pattern
of interest rates and house prices. These models are generic tools that
could be used in many different stress tests with different
specifications. The specifications actually define the ``stress'' in
the stress test. They include adjustments to reflect statutory
requirements, such as the requirement that default and severity rates
be ``reasonably related'' to the benchmark experience or that interest
rate increases greater than 50 percent reflect a correspondingly higher
rate of inflation. The specifications also include the house price and
residential rent paths and the interest rates that will apply during
the stress period. The data inputs to the models can change each time
the stress test is run. The data inputs include data on the
characteristics of loans owned or guaranteed by the Enterprises,
starting interest rates, and updated house and residential rent price
indexes, which are used to calculate current equity in the loan
collateral properties.
The general approach of the stress test to mortgage performance
involves three main steps: (1) estimation of statistical models of
mortgage performance (default, prepayment, and loan loss severity)
using Enterprise data covering a wide range of historical experience;
(2) adjustments to the statistical models to assure a reasonable
relationship to the benchmark loss experience; and (3) application of
the adjusted models to starting Enterprise mortgage portfolios in the
stress test. To assist the reader in understanding the more detailed
discussion of mortgage performance issues that follows, this section
provides a brief summary of some key issues concerning of the statutory
requirement to ``reasonably'' relate the performance of mortgages in
the stress test to the benchmark experience.
Because the benchmark sample contained only newly-originated,
fixed-rate, 30-year, owner-occupied, single family loans, the stress
test could not simply apply the rates of default and losses in the
benchmark loss experience and still take into account differences in
mortgage product types, seasoning of mortgages, and other factors the
Director considers appropriate, as required by the 1992 Act.\66\ Thus,
the first issue considered by OFHEO was how to link mortgage
performance in the stress test to the benchmark loss experience. The
primary question was whether to use a model-based approach to help link
the performance of an Enterprise's current loan portfolio to the
benchmark loss experience, or to rely upon a less sophisticated, but
less risk-sensitive approach. For reasons discussed under section III.
A. 3., Statistical Models of Mortgage Performance, OFHEO concluded that
the benefits of using a model-based approach exceed any potential
shortcomings.
---------------------------------------------------------------------------
\66\ 1992 Act, section 1361 (b)(1) (12 U.S.C. 4611 (b)(1)).
---------------------------------------------------------------------------
The next key issue was the choice of variables to include in any
statistical equations that would be part of a (statistical) model of
mortgage performance. OFHEO's choices in this regard were again
governed by the need to meet the multiple statutory objectives
described above, while also implementing a credit stress test based on
the historical benchmark loss experience. The stress test does not
project all differences in loan performance that may have been
identified in previous research. Rather, the factors used to project
mortgage performance are limited to those necessary to: (1) reflect
differences in characteristics of loans in implementing the credit risk
stress component of the stress test as required by the 1992 Act; and
(2) reflect differences in the interest rate environments experienced
by the loans in the stress test.
Other factors that relate to or explain differences in mortgage
performance are not, in OFHEO's view, appropriate to the proposed
regulation. Specifically, the stress test does not attempt to adjust
losses by incorporating factors to reflect changes in Enterprise
business practices subsequent to the benchmark loan origination and
loss experience.\67\
[[Page 18119]]
OFHEO believes that such adjustments would undermine the purpose and
intent of the statutory requirements to implement a credit stress test
based on the benchmark loss experience. In addition, although some
business practices that contributed to the losses of the past may have
been improved over time, a new severe economic environment may expose
other unobservable weaknesses. Furthermore, in reasonably relating
starting position loan portfolios to the ``experience'' of the
benchmark loans, it is not possible to separate the effects of business
practice from other aspects of the benchmark economic environment.
---------------------------------------------------------------------------
\67\ For example, both Enterprises have made changes to their
single family underwriting standards and practices since the time
the benchmark loans were originated in 1983-84, but no underwriting
variable is included. This particular issue is discussed in greater
detail below, in the context of comments received in response to
OFHEO's ANPR.
---------------------------------------------------------------------------
The proposed regulation also does not incorporate economic or
demographic variables that are not specifically prescribed for the
stress test, such as unemployment or divorce rates. Nor are such
variables included in the estimation of the statistical model used in
the stress test. If they were to be included, it would be necessary to
assume values for these factors in the stress period--values that are
consistent with the benchmark experience. Such an approach would
substantially increase the number of variables for which assumptions
would be required during the stress period, without gaining significant
value in predicting credit losses for Enterprise loan portfolios.
3. Statistical Models of Mortgage Performance
A threshold issue for OFHEO was whether to develop statistical
models of mortgage performance or to use a simpler approach, such as
applying a table of historical default, prepayment, and loss severity
rates.
a. ANPR Comments
Most of the comments related to this issue suggested that the
direct application of benchmark rates of default, prepayment and loss
severity would be problematic. A number of respondents to the ANPR
cautioned that direct application of benchmark default rates, which
were experienced during a period of declining interest rates, would not
be appropriate for the up-rate scenario of the stress test. Freddie Mac
suggested that OFHEO adjust benchmark default rates to the interest
rate environment or use a proportional downward adjustment to credit
losses. Mortgage Risk Assessment Corporation (MRAC) stated that it is
important to model the interaction between expected losses and expected
prepayments. America's Community Bankers (ACB) recommended joint
modeling of prepayments and defaults as the best way to capture
adjustments to housing values.
Fannie Mae, on the other hand, favored applying benchmark rates of
default and loss severity directly. More specifically, Fannie Mae
recommended that OFHEO model total loan terminations (defaults plus
prepayments) using a commonly applied method of relating total
terminations to interest rate movements (sometimes referred to as a
``total terminations model''). Fannie Mae recommended that the default
portion of total terminations should be based on observed default rates
for mortgages from the benchmark experience, with appropriate
distinctions based on different LTV ratios, mortgage product, and risk
categories. The level of prepayments would be calculated by subtracting
those defaults from total terminations. Fannie Mae stated that a
statistical model designed to predict defaults and prepayments
simultaneously would be difficult to replicate because it would employ
computer simulation methods based upon random numbers, known as Monte
Carlo simulations. Fannie Mae also expressed concern that the
Enterprises would have difficulty managing capital requirements based
on econometrically derived relationships, rather than on the certainty
of defined historical loss rates.
b. OFHEO Response
Based on its analysis of available information, including the ANPR
comments and relevant academic literature, OFHEO found that statistical
modeling has numerous advantages over alternative approaches, such as
applying tables of default, prepayment, and loss severity rates from
the benchmark experience.
First, statistical models are able to provide valid outcomes when
data inputs occur in different combinations from those observed in the
available historical data. This capability is important, because the
benchmark loss experience does not include large enough sample sizes
for all relevant loan products and risk classes to allow direct
application of benchmark loss rates to the Enterprises' starting loan
portfolios. Statistical models based on large samples of loans can
capture differential mortgage performance across a wide variety of
products and still allow the performance of each product to be related
to the benchmark experience. OFHEO has access to a rich database,
consisting of millions of detailed loan records from the Enterprises,
which allows for a statistical model of defaults and prepayments that
can capture the nuances of product distinctions.
Second, statistical models allow the stress test to extrapolate
reasonably to out-of-sample events, such as the sustained adverse
interest rate scenarios of the stress test.
Third, applying statistical models of mortgage performance provides
the ability to impose multiple statutory requirements in a logically
consistent manner. For example, the 1992 Act specifies rates of default
and losses in the stress test that are reasonably related to the
benchmark loss experience. The 1992 Act also provides that the Director
take into account the impact of ``mortgage seasoning'' and a variety of
other factors that delineate various mortgage product types (property
type, amortization type, amortization terms, ownership type, etc.).
Statistical models allow the stress test to address all these statutory
provisions when applying the two adverse stress test interest rate
scenarios.
OFHEO also found that using statistically derived models of
default, prepayment, and loss severity together with a cash flow
approach is the most accurate method to describe the financial
performance of the Enterprises on a monthly basis over the ten-year
stress period. Moreover, use of statistical models in the stress test
is consistent with the 1992 Act \68\ and the Congressional expectation
expressed in the House Report that the risk-based capital standard
``will be an economic model that will test the enterprises' financial
position under stressful economic situations.'' \69\ The House Report
also noted that:
---------------------------------------------------------------------------
\68\ The 1992 Act directs OFHEO to include in the regulation
``specific requirements, definitions, methods, variables, and
parameters used under the risk-based capital test.'' This direction
suggests that a statistical model was contemplated. The 1992 Act,
section 1361(e)(2) (12 U.S.C. 4611(e)(2)). Further, the Director is
required to ``provide copies of the statistical model or models'' to
other government agencies. 1992 Act, section 1361(f) (12 U.S.C.
4611(f)).
\69\ H.R. Rep. No. 102-206, at 62 (1991). See also, S. Rep. No.
102-282, at 24 (1992).
[t]he Department of the Treasury, the Congressional Budget Office,
the General Accounting Office, the Office of Management and Budget
and HUD have all stated that the proper way to ensure that Fannie
Mae and Freddie Mac have adequate capital is to use traditional
capital ratios in combination with sophisticated financial models,
or risk-based capital stress tests.\70\
---------------------------------------------------------------------------
\70\ H.R. Rep. No. 102-206, at 62 (1991).
Fannie Mae's recommendation to estimate a statistical model of
total terminations with default rates fixed at benchmark levels would
make it more difficult for the stress test to satisfy the
[[Page 18120]]
provisions of the 1992 Act that require OFHEO to consider seasoning and
the various loan characteristics described above. OFHEO is also
concerned that a model that derives prepayment rates as suggested by
Fannie Mae would not be consistent with section 1361(b)(2) of the 1992
Act, which directs that ``[c]haracteristics of the stress period other
than those specifically set forth in subsection (a), such as prepayment
experience . . ., will be those determined by the Director, on the
basis of available information, to be most consistent with the stress
period.'' The consistency of prepayment experience with the stress
period is best achieved by modeling both prepayment and default rates,
rather than using a statistical model of terminations with embedded
default rates that are not statistically determined.
OFHEO also found that the total terminations models to which Fannie
Mae refers are applied widely and usefully only in circumstances where
credit losses are not an issue (for example, in pricing mortgage-backed
securities for investors, where credit risk can be ignored because of
agency guarantees), or when the available data do not allow the analyst
to distinguish default terminations from voluntary prepayments (for
example, in the pool level data available from commercial sources).
This is not the case for the stress test.
OFHEO is sensitive to Fannie Mae's concern that a statistical model
of defaults and prepayments would be difficult to replicate. OFHEO does
not propose to base any component of the stress test on random number
(Monte Carlo) simulations. The model is straightforward and
transparent, so that it will be possible for the Enterprises to project
default and prepayment patterns in the stress period using their own
information about the composition of their business, and recent
economic trends.
As for complexity, OFHEO believes that there is no fundamental
difference in complexity between computing total termination rates from
the models mentioned by Fannie Mae, and computing them from the
separate default and prepayment rates generated by the model OFHEO has
proposed. Once the statistical model OFHEO proposes has been estimated
and calibrated, its application is no more difficult than the
application of a table of historical default rates. That is, the model
provides a means to ``look up'' the default or prepayment probabilities
for loans with a particular set of characteristics. Further, under the
approach proposed by Fannie Mae, the actual level of default rates
applied in the stress period would not actually be fixed, but would
vary with changes in the composition of an Enterprise's loan portfolio
and trends in property values that update borrower equity values. Under
either approach, determining the potential impact of market conditions
or changes in an Enterprise's portfolio on its capital requirement is
straightforward.
4. General Methodological Issues
A number of general issues arose in the context of using
statistical models to project mortgage performance in the stress test.
These issues required decisions about how to account for product
differences, what sources of historical data to use in estimating the
statistical models, and what level of data aggregation to use to
estimate and project mortgage performance. In addition, OFHEO received
a number of comments in response to ANPR questions on property
valuation issues. These were also considered in developing and applying
statistical models of mortgage performance. Each of these areas is
considered in the following sections.
a. Product Differences
The 1992 Act requires the stress test to capture both the unique
risk characteristics of various loan product and property types and
adjust for changing economics (house prices and interest rates) over
time. In deciding its approach to modeling default and prepayment
rates, OFHEO found it necessary to treat single family and multifamily
products separately because of the significant differences in
collateral property types and loan terms explained below.
The nature of the collateral property differs substantially between
single family and multifamily loans. Nearly all single family property
mortgages held by the Enterprises are owner-occupied.\71\ In contrast,
multifamily collateral produces income from rentals. Multifamily
mortgages are commercial loans on housing projects that compete for
market share among a very mobile population with short-term rental
contracts and relatively low moving costs. The household demographics
of apartment renters vary greatly from those of single family
homeowners and renters. The dynamics of construction cycles that
accentuate market booms and busts are also different for single family
and multifamily residences.
---------------------------------------------------------------------------
\71\ Even those that are rentals rely upon the performance of
one, or at most four, households.
---------------------------------------------------------------------------
Single family and multifamily mortgages generally have different
loan terms. In particular, to balance the desire of borrowers for
flexibility with the needs of investors for stability, multifamily
mortgages typically have ten- to fifteen-year balloon terms and initial
yield-maintenance periods of seven to ten years. During the yield-
maintenance period, borrowers may prepay, but they are subject to a
prepayment penalty until the maintenance period expires. Such
prepayment disincentives are not used in single family lending. Also,
in contrast to single family mortgages, multifamily mortgages tend to
be non-recourse, which means that multifamily lenders and guarantors,
have recourse only to the collateral, and not to the borrower's other
assets and income.
Because of these differences, OFHEO developed separate mortgage
termination models for single family and multifamily mortgages, with
all other property and product type differences handled as subsets of
these two primary classifications. This approach is consistent with
comments from HUD, Freddie Mac, ACB, and Mortgage Bankers Association
of America (MBA). However, there are many issues common to both the
multifamily and single family models, and the general modeling approach
to both models is similar in many respects.
In the ANPR, OFHEO solicited public comment on modeling approaches
generally and, more specifically, on how to relate the credit risk of
other loan product types to the 30-year fixed-rate mortgages used to
identify the benchmark experience. These comments are addressed below
in section III. A. 7., Relating Losses to the Benchmark Loss
Experience.
b. Historical Analysis Data
Another modeling issue faced by OFHEO was whether to use only
Enterprise data to estimate statistical models, or to use data from a
wider array of sources. A similar issue arose in the context of
identifying the benchmark loss experience. After considering ANPR
comments, OFHEO found that Enterprise data sets were the most relevant
sources currently available for determining a benchmark loss
experience, because Enterprise data is the most representative of the
experience of loans owned or guaranteed by the Enterprises. Further,
using Enterprise data is consistent with the general practice of
banking and thrift industry regulators and credit rating agencies,
which is to use data on the loss experience of comparable assets
[[Page 18121]]
for the relevant industry to determine credit quality and/or capital
adequacy.
For the same reasons, OFHEO also used Enterprise data to estimate
the statistical models for default and prepayment in the proposed
stress test. Using Enterprise data for this purpose provides
consistency between the estimates of the benchmark loss experience, the
estimation of the statistical models for default and prepayment, and
the aggregation of loan level data to create starting position data for
the stress test. It will also permit OFHEO to update the statistical
models over time, as needed, to capture new performance dynamics and/or
new products.
c. Aggregation
Another threshold issue for OFHEO was how to aggregate loan level
data to reduce the number of data records that must be stored and
processed, while preserving sufficient detail to capture differences in
loan performance among important risk classes in the stress test.
(i) ANPR Comments
MRAC stated that a loan level model would be most appropriate if
data were available, but a model that aggregates on the basis of the
origination year, loan term, coupon rate and current loan-to-value
ratio (CLTV) would be acceptable. Freddie Mac recommended that, if
OFHEO were to use a joint default/prepayment model, OFHEO should
construct a pool for each origination year, aggregated by mortgage
product, property type, occupancy status, and CLTV. Both MRAC and
Freddie Mac recommended that OFHEO not only aggregate data according to
CLTV, but also use CLTV as an explanatory variable in statistical
models of default and prepayment rates.
(ii) OFHEO Response
OFHEO proposes to aggregate single family loan level data into loan
groups based on the following characteristics: Enterprise, portfolio
type (securitized vs. retained), product type, origination year,
original LTV, original coupon, and region (Census division).
Multifamily loans are aggregated using the same categorical variables
as for single family loans, with an additional aggregation class for
original debt-coverage-ratio values. Single family loans purchased
during the stress period under existing contractual commitments are
grouped using all of the characteristics of existing loans plus month
of origination (representing the timing of delivery during the stress
period). All loan group records include additional fields for measured
characteristics, such as the total unpaid balance (UPB) for loans held
in portfolio, UPB-weighted average values for guarantee fees for
securitized loans, and original term-to-maturity.
OFHEO chose not to propose CLTV as a criterion for data
aggregation. Attempting to aggregate data by CLTV would be problematic
because CLTV value changes throughout the stress period. However, CLTV
is used to compute important explanatory variables used to predict
default, prepayment, and severity rates. These variables rely upon CLTV
to incorporate a loan seasoning process that updates property values at
the start of the stress test and then throughout the stress period.
d. Property Valuation
The 1992 Act requires that OFHEO take into account the impact of
the ``seasoning'' of mortgages on mortgage performance. As that term is
used in the statute, it requires accounting for changes in LTV due to
changes in housing values and the repayment of loan principal.
Accounting for changes in LTVs requires some method of updating
property values, in addition to computing scheduled amortization. The
first NPR proposed using the House Price Index (HPI), developed by
OFHEO, as the basis for updating single family housing values to meet
the statutory requirement for loan seasoning, in lieu of the Constant
Quality House Price Index published by the Secretary of Commerce.\72\
The HPI, which is published quarterly, provides average house price
appreciation rates for the nation, the 50 States and the District of
Columbia, and the nine Census divisions. It uses repeated observations
of housing values on individual single family residential properties.
These repeat observations arise where at least two primary mortgages on
the same property were purchased by either Freddie Mac or Fannie Mae
since January 1975.\73\ Index values are published starting with 1980.
---------------------------------------------------------------------------
\72\ 61 FR 29616, June 11, l996.
\73\ The procedures underlying the estimation of the HPI assume
that individual house price growth rates will be distributed around
the average growth rate through a log normal diffusion process.
---------------------------------------------------------------------------
In this NPR, OFHEO proposes the method by which loan seasoning will
be used to predict credit losses in the stress test, both for single
family and multifamily mortgages. For single family mortgages, the
OFHEO HPI is supplemented with various measures of the distribution of
individual house price growth paths around the average values measured
by the index. Three terms--dispersion, volatility, and diffusion--are
important concepts for understanding these measures and how the stress
test fulfills the statutory requirement that mortgage loans be
seasoned. ``Dispersion,'' refers to the distribution, at any point in
time, of the (cumulative) growth rates for values of each house in a
group, around the average growth rate for that group. Dispersion
results from ``volatility'' or variability of growth rate paths on
individual properties from the average growth rate path for all
properties. Volatility, like dispersion, can be measured through
statistical relationships. The underlying process by which a model
generates individual house price growth paths to yield various levels
of volatility and dispersion over time is called ``diffusion.''
Similar procedures are used to season multifamily loans, except
that there is no underlying property value index. Rather, property
value is estimated using indexes that first update property cash flows.
Still, the concepts of dispersion, volatility, and diffusion apply to
multifamily property values, and to the principal measures of borrower
equity in models of multifamily mortgage performance.
The ANPR posed several questions related to measurement of house
price dispersion and to the statistical validity of the HPI as a price
index. Issues raised by these questions will be discussed below.\74\
They are: the appropriate level of geographic aggregation for the HPI
in the stress test, how to account for the dispersion of house prices
around the mean in the loan seasoning process, and whether and how to
adjust for statistical biases and revision volatility inherent in the
HPI data and estimation methodology.\75\
---------------------------------------------------------------------------
\74\ The first NPR proposed the HPI as the index OFHEO would use
to season loans in the stress test, but did not address how OFHEO
would use that index in the stress test. Comments regarding the
first NPR will be addressed, together with comments on this NPR,
when OFHEO publishes a final Risk-Based Capital regulation.
\75\ ``Revision volatility'' refers to changes in previously
estimated index values that occur as a result of the addition to the
data of new repeat transaction pairs associated with current
transactions. Current transactions can change index values for prior
quarters, because every repeat sale of a property provides
additional information about house price changes during the time
since the prior transaction on that property.
---------------------------------------------------------------------------
(i) Geographic Aggregation
OFHEO's HPI is estimated at the level of individual States and the
nine Census divisions. A national index is also produced as a
population-weighted average of the nine Census division indexes.
Decisions regarding the level of geographic aggregation at which to
estimate and apply house price indexes
[[Page 18122]]
typically involve a tradeoff between the need to identify relatively
homogeneous market areas and the need for large enough samples of
repeat transactions to assure the accuracy of the indexes. This is,
simply put, a trade-off between the advantages and disadvantages of
creating indexes for smaller versus larger geographic areas.
At lower levels of geographic aggregation, both property types and
the local factors influencing house prices are more likely to be
similar, and therefore the average appreciation rate is likely to be
more representative of the trend in individual property values.
However, lower levels of geographic aggregation result in relatively
fewer observations for estimation, resulting in increased sampling
error in the estimated house price index.\76\ At larger levels of
geographic aggregation, the greater number of observations may yield
estimates of average price growth with smaller sampling errors, but at
the risk of not projecting accurately the appreciation rates of the
various submarkets.\77\
---------------------------------------------------------------------------
\76\ That is, if only a small number of repeat transactions are
available to calculate a price index, there is a greater chance that
the resulting index is not representative of price changes in the
particular housing market as a whole.
\77\ This situation could occur, for example, if two adjacent
smaller areas with different rates of appreciation are combined and
assigned the same average rate of appreciation through a common
price index. Whether this type of aggregation is ultimately a
problem depends on how the house price index is to be applied, and
whether it is to be applied to individual properties or to loan
aggregates.
---------------------------------------------------------------------------
(a) ANPR Comments
A number of comments were received on the issue of geographic
aggregation of house price indexes. All commenters implicitly
recognized the tradeoff involved in choosing the level of geographic
aggregation. The National Association of Realtors (NAR) recommended
using the lowest level of market aggregation possible, while at the
same time minimizing the variance of individual house prices in a
market area, and urged that the optimum level of aggregation be
determined by computational considerations. MRAC recommended that the
choice of aggregation level be driven by objective, external criteria,
such as minimizing estimation errors, and described its practice of
using the lowest level of geographic aggregation in constructing its
indexes, while using higher levels of aggregation for computing the
variances. Freddie Mac recommended that OFHEO use house price indexes
computed at the Census division level to avoid the need to rely on what
it called ``highly uncertain individual house-price volatility
processes'' that would be associated with the use of a national index
together with corresponding volatility measures. In addition, when
compared to State or local level house price indexes, Census division
level indexes would have lower standard errors and thus more reliable
predictions.
(b) OFHEO's Response
The choice of aggregation level of the HPI for the stress test is,
ultimately, a selection of the level that is most appropriate for the
seasoning of mortgages when estimating and projecting mortgage
performance. Because the stress test cannot determine the value of each
house securing every loan, some type of aggregation is needed. The
proposed stress test, therefore, combines estimates of average trends
in house prices with estimates of the dispersion of individual
appreciation rates around the average growth rate within a given
geographic area. This approach provides the maximum relevant
information about the equity position of borrowers.
After considering the alternatives and the comments, OFHEO believes
that using HPI indexes computed at the Census division level combined
with estimates of dispersion of individual appreciation rates around
the divisional indexes would be appropriate. OFHEO found that available
data is not sufficient to generate statistically valid State-level
indexes for some of the less populous States. OFHEO has not proposed to
use indexes below the State level (at the metropolitan statistical area
(MSA) level, for example), because there are too few areas in which
statistically valid indexes can be estimated.
OFHEO agrees with Freddie Mac's comment that Census division
indexes without volatility measures reflect regional dispersion better
than using a national index with such measures. While OFHEO does
publish State-level HPI series, these series are not statistically
valid for some of the less populated States. Using Census division
indexes, in combination with estimates of individual house price
volatility and the resulting dispersion in each division, provides a
more complete characterization of housing value dynamics both within
and across regions.
MRAC's practice of using a larger level of geographic aggregation
for volatility estimates than is used for the price index itself is
appropriate when price indexes are based on very small aggregation
levels, for example, at the MSA level. Using a larger area to measure
volatility helps to diminish the small sample problems of generating
price indexes for very localized markets. However, the same is not true
when estimating price indexes at the Census division level, because
there are no small-sample problems at that level of aggregation.
Furthermore, applying national level volatility to division-level price
indexes would defeat the purpose of using the division-level indexes.
National volatility measures of individual house price growth could be
so large that divisional variations in average house price growth
become meaningless.
(ii) Volatility and Diffusion
Choosing to use Census division level price indexes with dispersion
measures opens additional issues. In particular, capturing the
dispersion of house price growth rates around an index value requires
both a measure of volatility and a particular diffusion process to
translate volatility into actual dispersion. Several ANPR commenters
addressed these issues in the context of their discussions of
geographic aggregation.
(a) ANPR Comments
Comments received in response to the ANPR differed on whether and
how to estimate the dispersion of individual house-price-appreciation
rates around the average rates implied by a house price index. Both
MRAC and the Department of Veterans Affairs (VA) recommended that OFHEO
use a stochastic (random) diffusion process to allow volatility
measures to generate a normal (bell-shaped) distribution of individual
house prices around the mean prices implied by index values. MRAC noted
that failure to do so would underestimate dispersion, even if a highly
disaggregated index were used. MRAC observed that underestimation of
dispersion could cause underestimation of default and severity rates.
MRAC also stated that the tradeoff between the accuracy of the larger
sample size and the greater geographic specificity of a smaller sample
is even more important in estimating the variance (volatility) than in
constructing the index.
Both Fannie Mae and Freddie Mac, on the other hand, recommended
against using a stochastic process to estimate dispersion of house
values. Freddie Mac argued that one cannot directly observe the
volatility of house-price growth rates, and that attempts to estimate
it have thus far failed to achieve adequate consistency. Nor is it
necessary to estimate volatility, Freddie Mac argued, because the
variation in house price indexes across Census divisions
[[Page 18123]]
captures a significant amount of the house price dispersion around a
national house price index, as well as the basic shape of the house
price distribution for Enterprise loans.
Freddie Mac also questioned OFHEO's assertion in the ANPR that
dispersion increases over time. It suggested that models that impose
increasing dispersion on house price changes, such as ``random walk''
models, are inappropriate because long-run market forces keep the
appreciation of individual houses moving roughly with the national
average, and because the data do not support such models. Freddie Mac
asserted that such models systematically overstate dispersion for
longer holding periods and could significantly and artificially inflate
the capital requirement.
(b) OFHEO's Response
OFHEO understands the reason for Freddie Mac's concerns about
volatility, but notes that Freddie Mac's comments preceded OFHEO's
first publication of the HPI. Based on its experience in estimating the
HPI, OFHEO now finds it possible to estimate house-price volatility
with adequate reliability, particularly for indexes estimated at the
Census division level. Volatility measures are produced as part of the
statistical process used to generate the OFHEO HPI. These measures are
used to summarize the underlying diffusion process and characteristic
dispersion of house price growth paths as a function of time. The
volatility measures (parameters) are published in the OFHEO HPI Report.
They model dispersion as a function of mortgage age. OFHEO preferred
such a stable process to one that relies on stochastic processes that
yield different results every time they are used. Because the OFHEO HPI
volatility parameters are produced with the HPI itself, they provide
results consistent with the HPI, and they are, therefore, OFHEO's
choice for capturing house price dispersion in the proposed stress
test. However, OFHEO agrees with Freddie Mac's concern that estimates
of dispersion for longer holding periods may be unreliable, and has
adopted an approach in which estimated dispersion is held at fixed
levels after mortgages reach a certain age.\78\
---------------------------------------------------------------------------
\78\ This age varies by Census division, but is approximately 15
years from mortgage origination. The formula for computing the
maximum allowable age for each Census division can be found in
section 3.5.2.3.2.3., Probability of Negative Equity
(PNEQq), of the Regulation Appendix.
---------------------------------------------------------------------------
(iii) Revision Volatility
Revision volatility primarily affects growth rate estimates for the
most recent quarters included in the index. This is due to the fact
that relatively more additional data is added affecting these quarters
than earlier quarters.
(a) ANPR Comments
OFHEO received a number of comments in response to the ANPR on
whether changes in the index resulting from revision volatility should
be reflected in the stress test and, if so, with what frequency. NAR
suggested that revisions should be made at the same time OFHEO is
required to re-estimate the capital standards. In contrast, MRAC
suggested using a ``chaining method'' \79\ that precludes the need for
revision to index values for historical periods. The chaining method
eliminates revision volatility because it does not revise data of
earlier periods as new data become available. Freddie Mac suggested
that OFHEO calculate the revisions so as to exploit the greatest
possible set of information, but moderate the resulting volatility of
the capital requirement by placing limits on the size of the quarterly
or annual revisions to the indexes. ACB argued for a reasonable advance
notice to the Enterprises prior to any changes in the capital
requirement resulting from changes in the indexes to enable them to
engage in reasonable business planning.
---------------------------------------------------------------------------
\79\ The chaining method involves the following steps: (1)
estimation of a historical reference index using all repeat
transactions data available as of a specified date, after which no
revisions in previously estimated index numbers will occur; (2)
acquisition of new data providing information on the most recent
time period, and including additional repeat transactions that pair
with transactions in previous periods; (3) application of the most
recently updated index series to inflate the first property value
for a repeat transaction pair to update this value to the
penultimate (next-to-last) time period; and (4) estimation of the
index number for the last time period using the pseudo-repeat
transactions data created in steps (1)-(3).
---------------------------------------------------------------------------
(b) OFHEO's Response
The proposed stress test does not include an adjustment for
revision volatility. Since the time the issue of revision volatility
was raised in the ANPR, OFHEO has determined that revision volatility
is not likely to have a significant impact on risk-based capital.
Revision volatility primarily affects growth rate estimates of the most
recent quarters, which will be those immediately preceding the start of
the stress test. For loans that have been outstanding for several years
at the start of the stress test, changes in appreciation rates in the
most recent quarters will represent a small proportion of the total
change in housing values since origination. For loans that have been
outstanding only a short time at the start of the stress test,
projected changes in house prices and in LTV will be minimal in any
case, due to the fact that little time has elapsed since origination,
and quarter-by-quarter appreciation rates are generally small.
Consequently, OFHEO does not expect revision volatility to affect risk-
based capital requirements. OFHEO also proposes not to revise the house
price index used to determine the appreciation rates applied in the
stress period. Rather, HPI values, as published in the 1996, third
quarter, HPI Report, will be the basis for relating stress test
economic conditions to the benchmark experience.
OFHEO chose not to propose the chaining method suggested by MRAC
because it fails to use all of the available data in estimation. In
particular, the chaining method uses information on recent property and
mortgage transactions only for calculating appreciation rates in the
most recent period, ignoring the information provided by these
transactions on appreciation rates in earlier periods.
(iv) Statistical Biases
In the ANPR, OFHEO requested comment on whether the HPI should
include adjustments for identifiable sources of statistical bias, on
how sample selection bias should be addressed,\80\ on whether a
statistical adjustment should be made to address appraisal bias,\81\
and on what additional sources of statistical bias exist and how they
might be addressed. In NPR1, OFHEO stated that it would make no
[[Page 18124]]
adjustments to the HPI itself, but would discuss in the second NPR
whether such adjustments were to be made in the stress test.
---------------------------------------------------------------------------
\80\ Sample selection bias refers to the possibility that using
repeat transactions as the selection criteria, rather than random
selection, could result in an index that is biased. Selection bias
results when the probability that a property does or does not repeat
is correlated with the change in value. For example, bias can result
when the period between transactions is correlated with the change
in house prices. Because more rapidly appreciating properties turn
over within shorter time intervals, they are more likely to appear
in the sample used for estimation. In addition, properties that are
sold or refinanced are likely to be the ones that have had higher
than average appreciation.
\81\ Appraisal bias can result from the perceived tendency of
appraisers, as agents of primary mortgage lenders, to impart an
upward bias to a home value to insure that a home sale is made.
Appraisal bias also occurs when the use of appraisals to value
property at refinancing may smooth the fluctuations in housing
values because appraisals are derived from comparisons with
properties that have either been sold or listed for sale within the
past several months and may fail to indicate more recent changes in
housing value. In fact, listings are only used in case circumstances
where actual sales are few and far between, most often in rural
areas.
---------------------------------------------------------------------------
(a) ANPR Comments
As a general comment, Freddie Mac cautioned that research on
potential sources of bias is relatively new and that attempting to
``un-bias'' future price index values estimates introduces a high
degree of complexity. Consequently, Freddie Mac recommended keeping the
house price index simple until research on potential bias is more
conclusive. Freddie Mac also suggested that the reliance of the
weighted repeat sales technique on the ordinary least squares (OLS)
method \82\ may result in bias because that methodology does not
generally provide robust estimates of central tendencies in the
presence of outlier observations, where appreciation is especially
large or small. Freddie Mac suggested eliminating outliers or ``down-
weighting'' them, for example, by using a median regression.
---------------------------------------------------------------------------
\82\ Ordinary least squares is the most commonly used
statistical technique for simultaneously analyzing the relationship
of many explanatory variables to one special variable of interest
(called the ``dependent'' variable).
---------------------------------------------------------------------------
(b) OFHEO's Response
OFHEO agrees with Freddie Mac that attempts to adjust the HPI would
be premature and should await more conclusive research. OFHEO also
agrees with Freddie Mac's general observation on the sensitivity of OLS
estimates to outliers, but has concluded that adopting another
estimation methodology is unwarranted. It should be noted that the
weighted-repeat sales (WRS) methodology \83\ applied to estimate the
OFHEO HPI uses information obtained from a first-stage OLS estimation
to develop weights that have the effect of discounting the impact of
transactions that occur far apart in time. Because these are the
transactions that are presumed under the WRS method to have the largest
sampling variability, and therefore those most likely to contribute
outliers, the WRS method automatically accounts for the potential
impact of outliers. In addition, OFHEO reports median rather than mean
appreciation rates, which diminishes any potential impact of outlier
data.\84\
---------------------------------------------------------------------------
\83\ This methodology, which is explained in the first NPR, uses
pairs of transactions (i.e., repeat sales) involving the same homes
to estimate home price appreciation.
\84\ The WRS methodology used to generate the OFHEO HPI actually
computes median growth rates, directly. These rates need to be
adjusted to compute mean growth rates. In NPR1, these were referred
to as geometric and arithmetic means, respectively.
---------------------------------------------------------------------------
(v) Sample Selection Bias
Repeat-sales and repeat-transaction price indexes do not include
property value information from all mortgage transactions. Issues of
potential bias in the measured house price appreciation rates arise
because the sample of properties on which repeated transactions are
available may not be fully representative of all properties in a given
market area.
(a) ANPR Comments
A number of comments were received on sample selection bias in
generating a house price index. Freddie Mac noted that sample selection
bias results from using only properties that have been sold or
refinanced. The selection of these properties is not random and is
correlated positively with price appreciation. That is, properties with
lower rates of appreciation will have fewer sales and refinancings, and
thus provide relatively fewer observations for calculation of the HPI.
Although Freddie Mac recommended that this issue be addressed by using
a WRS index, which provides retrospective information by pairing two
transactions on the same property at different time periods, it noted
that some sample selection bias is present in the near term.
NAR suggested that sample selection bias results from the movement
of an individual property from government mortgage insurance programs
(Federal Housing Administration (FHA) VA) into the conforming
conventional market, and vice versa, because the lower property values
captured in the government insurance and guaranty programs might not be
matched in the WRS series. If price appreciation in a market area is
distributed unevenly with respect to selling price (i.e., lower priced
homes appreciate slower or faster than do higher priced homes), the
absence of a match at the lower end may introduce a bias in the level
of price appreciation for the market under evaluation. NAR suggested
that using FHA data, to the extent it is available, to construct the
weighted repeat sales transactions, would adjust for the low-end sample
selection bias. NAR also suggested that OFHEO investigate using
different criteria with respect to time between repeat transactions
entering the Enterprise loan history file to determine if the end of
sample bias is significant, and to possibly suggest ways of correcting
for it. NAR suggested that one way of correcting for any such bias
would be to restrict the repeat sales in the sample to three-, five-,
and seven-year matches and to evaluate the level of bias that results.
ACB suggested that the effect of sample selection bias resulting
from the tendency to have greater turnover in that part of the housing
stock in which price appreciation has been stronger could be determined
by a separate analysis of the relationship between a foreclosure
property index and the overall price index. MRAC suggested that some
bias might result from properties leaving the sample because they have
appreciated enough that the size of subsequent mortgages on those
properties is above the conforming loan limit. MRAC then suggested that
indexes built on Enterprise data be compared to other more broadly
constructed indexes, such as those estimated by MRAC, that include all
properties that initially meet the conforming limit. MRAC also
suggested that the incidence of default and expected losses would be
underestimated if the impact of junior liens were not taken into
account.
(b) OFHEO's Response
OFHEO believes that no adjustments are necessary to correct for
potential sample selection bias. Low-end sample selection bias due to
the exclusion of FHA loans should not have a significant impact on the
HPI. FHA loans do not represent the entire lower end of housing
markets. There is ample representation of lower valued loans and
properties in the data used to estimate the HPI, in part because the
Enterprises promote affordable lending and are subject to HUD
affordable lending regulations. Furthermore, although FHA eligibility
requirements have historically been less restrictive than conventional
lending requirements, current trends in conventional lending are toward
more flexible standards, including lower down-payment requirements.
Although OFHEO agrees with MRAC that the conforming loan limit may
itself produce some bias in repeat transactions index values, this bias
is not significant in the HPI. Bias resulting from the conforming loan
limit would occur in high-cost housing markets where there are
significant numbers of homes with values near the conforming loan
limit, and where appreciation rates are greater than the national
average. As home values and loan amounts increase in these areas, new
loans may no longer be eligible for purchase by the Enterprises, and
the property appreciation cannot be captured in the HPI. However, such
bias would occur only in very isolated instances. First, the conforming
loan limit is substantially
[[Page 18125]]
above the average home price in nearly all areas of the country. The
loan limit would only create a significant issue for the stress test if
OFHEO were to use State, rather than Census division, indexes. The
potential in particular States with high-cost metropolitan areas for
sample selection bias resulting from the conforming loan limit becomes
less relevant when the HPI is estimated at the Census division level.
Second, the loan limit is updated annually by a factor representing
national house price appreciation.\85\ Third, borrowers may obtain two
mortgages on a property in order to take advantage of the interest rate
advantages of having a first mortgage under the conforming limit. In
that situation, repeat transactions are captured by the HPI even if the
total amount of mortgages on a property exceeds the conforming loan
limit. All of these factors suggest that the conforming loan limit is
not a significant source of bias in the OFHEO HPI.
---------------------------------------------------------------------------
\85\ The conforming loan limit is administered by the Federal
Housing Finance Board.
---------------------------------------------------------------------------
(vi) Appraisal Bias
Because interest rates have generally fallen since the early
1980's, most of the mortgage transactions used in estimating the HPI
are refinancings, rather than loans for home purchase. This fact raises
the question of the consistency between actual prices recorded on
purchase-money mortgages and appraisals used for refinance mortgages.
(a) ANPR Comments
Several comments on appraisal bias were received. Freddie Mac
recommended against using a statistical adjustment to the HPI to
address the impact of appraisal bias, asserting that it is far from
clear whether indexes based solely on purchase prices, versus those
based on a combination of purchase prices and appraisal values, better
represent true house-price appreciation rates. Freddie Mac asserted
that the common notion that purchase price is the ``true'' price is a
misconception, since the purchase price is but one of a distribution of
potential prices for any given house at any time. In light of the
current uncertainty over the extent of the bias, Freddie Mac believes
that it would be premature for OFHEO to attempt to develop a model to
correct for it.
MRAC suggested that eliminating transactions in which an appraised
value is used for either ``sale amount'' in the matched pairs would be
desirable, but may not be practical. MRAC cited its own research to
suggest that appraisal bias causes the yearly price appreciation
measured by transaction-based indexes to be one percentage point too
high. ACB suggested that construction of house price indexes with and
without refinance transactions would permit an assessment of about
whether appraisal bias is a significant phenomenon.
(b) OFHEO's Response
OFHEO agrees with Freddie Mac's recommendation that adjustments in
the HPI for potential appraisal bias not be made. Issues of statistical
bias merit further research and analysis, but at the present time OFHEO
is aware of no better alternative index to use in the stress test.
Also, measuring HPI only on actual purchase prices would compromise the
statistical reliability of the indexes over time, because the majority
of property values used in generating the various HPI indexes come from
refinancing transactions, using appraisal values.
In response to MRAC's comment on appraisal bias in appreciation
rates, it should be noted that the mere existence of identifiable
differences due to use of appraisals does not outweigh the overall
benefit of using the HPI in the stress test. Further, it is unlikely
that any appraisal bias that may exist in the HPI would have a
meaningful effect on risk-based capital because of the way in which the
HPI is used in the stress test. The mortgage performance models in the
stress test rely upon statistical equations that relate explanatory
variables developed using the historical HPI to actual, historical
mortgage performance. The same historical HPI series is used to season
(update LTVs of) existing loans to the start of the stress period.
Using the same HPI series to estimate the statistical model and to run
the stress test eliminates the effect of any appraisal bias in the HPI
on default and prepayment rates in the stress test.\86\
---------------------------------------------------------------------------
\86\ Appraisal bias could, theoretically, affect the rates
generated by the stress test if the method of computing the HPI were
changed in some way to account for appraisal bias or if appraisal
bias were found to be significantly different in more recent data
than in the historical data used to estimate the models. OFHEO does
not believe the change in the amount of appraisal bias in the HPI,
if any, is significant.
---------------------------------------------------------------------------
(vii) Multifamily Loans
For multifamily loans, OFHEO does not propose to use the HPI or any
other repeat-sales or repeat-transaction index to update property
values. There is not enough data available for OFHEO to develop its own
price index, and the only known price indexes blend many commercial
property types, have small numbers of observations, and are national in
scope. To overcome these data problems, OFHEO proposes to use an
earnings-based method for updating property values.
Multifamily loans are commercial loans for which property value
depends upon the stream of earnings generated by the property. For
these loans, OFHEO proposes to base the property value on earnings
multiplied by a price-to-earnings capitalization factor. The
capitalization factor summarizes the present value of a stream of
expected future earnings for a given property, using current interest
rates at each month of the stress test to discount the expected
earnings stream. Earnings are a function of net operating income at
loan origination, rental inflation, and the change in vacancy rates
since loan origination. The proposed stress test updates the price-to-
earnings capitalization factors as a function of changes in interest
rates, holding property-specific characteristics constant. In this way,
the stress test updates property values and seasons multifamily loans
in the proposed stress test.
In choosing the actual rent growth and vacancy indexes used to
update property earnings over time, OFHEO used government data where
available. Government data were available for all statistical analysis,
and for seasoning loans to the start of the stress test. In particular,
the model performs the statistical analysis and the seasoning of
existing loans to the start of the stress test using the rental cost
component of the Bureau of Labor Statistics Consumer Price Index (CPI)
to create a geographic specific rent index. Vacancy rates are not
needed for pre-stress period seasoning, but are used in estimating the
statistical model. The series used is the rental property vacancy
series published by the Bureau of the Census (Census Vacancy
Series).\87\ Because Enterprise purchases of multifamily loans are
heavily concentrated in MSAs, MSA indexes are used, where available, to
update property values.
---------------------------------------------------------------------------
\87\ The CPI and Census Vacancy Series are both based on single
and multifamily rental properties. OFHEO believes that the inclusion
of single family rental properties in the samples used to calculate
vacancy rate and rent growth rate series is not a serious concern
for the stress test. These series capture the cyclical dynamics of
multifamily rental markets, and are useful for updating property
values before and during the stress period.
---------------------------------------------------------------------------
Government data are not available for the entire stress period
itself. As explained later in the discussion under section III. A.7.,
Relating Losses to the Benchmark Loss Experience, the stress
[[Page 18126]]
test links stress period losses to the benchmark experience in part by
specifying benchmark rates of property value appreciation. However, CPI
rental cost data is not available for the benchmark time and place, and
Census Vacancy Series rates are only available for the benchmark
experience starting in 1986. To deal with this absence of government
data, OFHEO created a rent index consistent with the CPI data, but
based upon apartment data available from the Institute for Real Estate
Management (IREM). To fill in benchmark experience vacancy rates for
1984-1985, OFHEO also used IREM vacancy data to estimate the Census
Vacancy Series. The estimated government series are consistent with the
data used to estimate the mortgage performance models and season the
loans prior to and during the stress period itself.
Volatility estimates for rental rate inflation and vacancy rates
are used to calculate the dispersion of multifamily property values, in
much the same way volatility measures for the HPI series are used to
measure dispersion of property values for single family loans.
5. Default/Prepayment Issues
a. Use of Conditional Default and Prepayment Rates
A threshold issue for OFHEO was whether to construct statistical
models of conditional rates of loan defaults and prepayments or to
adopt a less detailed approach, such as calculating only cumulative
rates and distributing them in fixed percentages across the ten years
of the stress test. A conditional rate of default or prepayment refers
to the volume of loans that default or prepay during any period,
expressed as a percentage of the total volume of loans surviving at the
start of that period. The term ``surviving loans'' means those from the
group that have not previously prepaid or defaulted. A cumulative rate
of default or prepayment is the total percentage of a group of loans
that default or prepay during the entire period being studied (such as
the ten-year stress period). A group of loans studied over a ten-year
period would have a single cumulative default rate, but would have ten
annual conditional default rates.
(i) ANPR Comments
The ANPR asked whether default rates should be expressed in terms
of conditional failure rates, cumulative default rates, or in some
other manner. In response, MRAC stated that ``[d]efault rates are best
measured by cumulative life-of-loan rates with conditional rates for
each time period determined by estimating `seasoning curves' similar to
the Standard Default Assumption of the Public Securities Association
(PSA) \88\.'' ACB's comments, which emphasized the importance of
modeling the shrinking population of loans exposed to the credit risk
in the declining rate scenario, assumed that a conditional rate
approach should be used. Similarly, a preference for conditional rates
of default and prepayment is also implicit in NAR's assertion that the
principal merit of using a joint default/prepayment model is that it is
capable of using all available information to determine whether a
mortgage survives from one year to the next.
---------------------------------------------------------------------------
\88\ PSA has subsequently changed its name to the Bond Market
Association. The PSA Standard Default Assumption is to allow monthly
conditional rates to increase from zero to some peak rate over the
first 30 months of mortgage life, to hold that peak rate constant
for another 30 months, and then to allow monthly rates to decline
for an additional 60 months. The final rate reached at the end of
120 months is held constant throughout the remaining life of the
loans (Public Securities Association, Standard Formulas for the
Analysis of Mortgage-Backed Securities and Other Related Securities.
New York: Public Securities Association, update No.7, June 29, 1993,
at SF-14.).
---------------------------------------------------------------------------
Freddie Mac and Fannie Mae, however, recommended using cumulative
default rates to simplify the analysis. Freddie Mac was concerned that
conditional prepayment rates would lead to absurdly high default rates
in an up-rate stress test. In the up-rate scenario, prepayment rates
would be low, more loans would be outstanding, and default rates
conditioned on the number of loans outstanding would result in more
defaults. Freddie Mac recommended using actual cumulative default rates
from the worst region, which, implicitly, would include the same
prepayment effect as that which occurred during the benchmark period.
(ii) OFHEO Response
OFHEO proposes to apply statistical models of conditional rates of
default and prepayment for both single family and multifamily mortgages
in the stress test. The advantages of this approach are numerous. The
proposed approach automatically accounts for the impact of defaults on
the number of loans remaining active and subject to the risk of
prepayment, and vice versa. This feature is essential to develop a
reasonable representation of Enterprise mortgage cash flows across the
different economic scenarios envisioned by the stress test. It also
avoids potential numerical anomalies that might arise when total or
annual defaults during the stress test are fixed, such as years in
which total defaults would exceed total surviving loans due to high
prepayment levels in the declining-rate scenario of the stress test.
Also, the periodic nature of mortgage payments, scheduled amortization,
and the coupon adjustments on adjustable rate loans, all of which
affect mortgage performance, require a model that reflects a discrete
time period for each default or prepayment event.
OFHEO believes that a statistical model of conditional defaults and
prepayments is more accurate and more sensitive to stress test economic
factors, and to the Enterprises' starting books of business, than are
simpler methods that might be developed. Each quarter the test is
applied, a statistical model can account for changes in economic
conditions (such as the level and shape of the Treasury yield curve or
recent trends in house prices) and the composition of an Enterprise's
business since the last time the test was performed. That is, the rates
of default and prepayment applied when the stress test is run are
adjusted to reflect current circumstances. Such adjustments are
particularly important because mortgage prepayment and default rates
are highly time-dependent, characteristically increasing during the
first years following origination, peaking sometime between the fourth
and seventh years, and declining over the remaining years. However,
this time-characteristic pattern is itself affected by economic
conditions.
Another advantage of modeling conditional default and prepayment
rates is the support this approach provides for the proper treatment of
loss severity. Loss severity is affected significantly by factors that
affect the timing and amount of defaults in the stress test. Loss of
loan principal balance, the single largest cost element in determining
loss severity, is dependent upon house price declines, which are
dependent upon economic conditions leading up to the date of default.
Funding costs are also affected by the changing interest rates in the
stress test, as explained in later discussions under section III. A.
6., Loss Severity. For all of these reasons, using conditional default
and prepayment rates during each month of the stress period greatly
improves the sensitivity of the stress test to risk factors.
The proposed approach is, overall, responsive to concerns raised in
the ANPR comments, although OFHEO has proposed models of conditional
rates of default and prepayment, rather than accept the recommendation
of several commenters to use cumulative rates. NAR and ACB recommended
use of
[[Page 18127]]
conditional rates. As ACB recognized, the stress test must account for
the shrinking population of loans exposed to credit risk in the
declining rate scenario. Only through the application of conditional
default and prepayment rates is it possible to account for this
shrinking population under the alternative interest rate scenarios of
the stress test.
MRAC recommended measuring cumulative life-of-loan rates with
conditional rates for each time period determined by estimating
``seasoning curves'' similar to the Standard Default Assumption of the
Public Securities Association to determine conditional rates. OFHEO
proposes a model with much the same features suggested by MRAC. This
model uses mortgage age in the statistical default equations to provide
a baseline default rate time-series analogous to the PSA assumption.
(See note 41, infra.) That baseline is scaled, or multiplied upward, in
the same way that PSA recommends using its baseline curve, when the
stress test adjusts or ``calibrates'' its statistical default equations
to relate them to the benchmark experience. (See section III. A. 7.,
Relating Losses to the Benchmark Loss Experience.)
OFHEO's approach is also responsive to the recommendations of
Fannie Mae and Freddie Mac to keep the models simple. OFHEO proposes to
minimize the number of explanatory variables and to create as much
consistency as possible across different mortgage types while still
capturing differential credit risk by mortgage type. The models are
also ``simple'' in that the mortgage performance equations used in the
stress test can be used by the Enterprises--without any modifications-
to replicate the stress test. Further, OFHEO believes that using
cumulative default rates would not achieve significant simplification.
Freddie Mac's comments recognized that default and prepayment rates are
not uniform among loans with different characteristics. To deal with
these important differences, Freddie Mac suggested developing a system
of multiples and LTV categories that would be applied to historical
cumulative default rates. However, this approach requires a matrix of
rates that becomes, in practice, more complicated to estimate than a
statistical model of conditional default rates. Therefore, developing a
statistical model, based upon well-recognized techniques that are
widely used in the mortgage industry, was, in OFHEO's view, a
preferable approach.
b. Identifying Events for Default and Prepayment
A practical issue for modeling default and prepayment rates is how
to identify a default or prepayment event in the historical Enterprise
data.
(i) ANPR Comments
A number of ANPR commenters, including MBA and Freddie Mac,
suggested defining default events only in terms of foreclosures,
because many delinquencies are cured and do not generate significant
losses. In contrast, the VA suggested modeling the timing of cash flows
associated with all delinquencies, including loans that are reinstated
and do not terminate.
Only Freddie Mac addressed the subject of curtailments as a form of
prepayment. Curtailments are partial prepayments, made in addition to
regularly scheduled mortgage payments. Freddie Mac did not suggest that
they be tracked as mortgage events, but only that some consideration of
them be given in the calculation of current LTV ratios to account for
the resulting improvements in borrower equity positions. Freddie Mac
cited a study on Ginnie Mae curtailment speeds,\89\ and suggested that
Enterprise loan pools might have higher rates of curtailment than found
in the study, because of better borrower equity and liquidity
positions.
---------------------------------------------------------------------------
\89\ Peter Chinloy, ``Elective Mortgage Prepayment: Termination
and Curtailment,'' Journal of the American Real Estate and Urban
Economics Association 21 (3, Fall 1993), 313-332.
---------------------------------------------------------------------------
(ii) OFHEO Response
OFHEO agrees with MBA and Freddie Mac that the stress test should
not consider all delinquencies to be defaults. Only delinquencies that
result in termination of the loan are treated as defaults in the stress
test. Historically, these events predominantly have been foreclosures,
although today these events also include pre-foreclosure sales, where
delinquent borrowers sell their properties before foreclosure and share
the losses with the Enterprise and/or mortgage insurer.\90\ OFHEO found
that the more detailed modeling of delinquencies suggested by the VA
would make the model more complex and would not have a significant
impact on risk-based capital. The impact would be minimal, because in
the time and place of the benchmark loss experience, few, if any,
alternatives to foreclosure were utilized by the Enterprises and the
benchmark rates would, therefore, not change. Also, even if modest
improvements to the stress test were possible by modeling delinquency
events, at this time there are insufficient data to support an analysis
of delinquency resolutions and costs.
---------------------------------------------------------------------------
\90\ A less important default termination event is the transfer
of the property deed, in lieu of foreclosure. This is a foreclosure-
like event in that it results in the Enterprise taking title to the
property and having to manage and sell it, just as is the case with
foreclosed properties.
---------------------------------------------------------------------------
Mortgage default and prepayment events result from a borrower's
decision to terminate the mortgage, either by prepaying or defaulting,
resulting in an observed last-paid installment, after which no further
payments are forthcoming. In the case of (full) mortgage prepayment,
the borrower terminates the loan by repaying the remaining principal
and any outstanding interest. The models identify prepayment events in
the Enterprise data by the existence of a last-paid installment date
and a change in the loan status from active to prepay. Loan defaults
are identified as any loan that has terminated without an indication
that it has been prepaid or paid off at maturity.
In the proposed stress test, curtailments made prior to the
beginning of the stress period are accounted for in the starting loan
balances reported to OFHEO from the Enterprises. OFHEO does not,
however, propose giving further consideration for potential
curtailments in the stress period itself. OFHEO has found no evidence
that curtailments have a significant impact on current LTVs of
Enterprise loans on a portfolio-wide basis.\91\
---------------------------------------------------------------------------
\91\ The Chinloy study cited by Freddie Mac, which used a
limited data set, found that curtailments in the study period
(January 1988-May 1989) amounted to a very small rate (0.42 percent
per year) on the outstanding loan balances of the Ginnie Mae
security pools. Ibid., p. 326. More recent work by Fu, Lacour-
Little, and Vandell, on conventional mortgage curtailment rates,
also shows that curtailments amount to a small percentage of
portfolio balances. Qiang Fu, Michael Lacour-Little, and Kerry
Vandell, ``Retiring Early: an Empirical Analysis of the Mortgage
Curtailment Decision,'' unpublished manuscript, University of
Wisconsin--Madison, December 1997. These authors observed 25,566
mortgages for a 21-month period. These included a mixture of
conforming and jumbo loans, and included loans originated from 1967
to 1995. During a 21-month observation period, these authors found
that over 86 percent of the loans surveyed made no curtailments, and
only 0.64 percent of the loans made curtailments in excess of one
percent of the original loan balance. Ibid, Table 3, p. 22. The
largest curtailments were made on older loans (close to 20 years
old), where loan balances and default rates will be small to begin
with. Thus, any effect of these curtailments on credit losses would
be insignificant for risk-based capital determination.
---------------------------------------------------------------------------
c. Use of Joint Default/Prepayment Models
A key issue raised in the ANPR was whether to use a joint
prepayment and
[[Page 18128]]
default model or some simpler assumptions about default and prepayment
rates in the stress test. In the ANPR, OFHEO also asked whether
prepayments during the stress test should affect the volume or timing
of defaults.
(i) ANPR Comments
Several commenters supported the use of a joint model of defaults
and prepayments. MRAC stated that the ``absolute merits'' of the
approach are ``obvious.'' NAR asserted that the principal merit of
using a joint model of conditional default and prepayment probabilities
is its ability to use all the available information to determine
whether a mortgage survives from one year to the next or is lost from
the portfolio through prepayment or default. HUD cited the need to
model defaults and prepayments together as simultaneous decisions based
on the underlying property equity.
The Enterprises opposed a joint default and prepayment model.
However, Fannie Mae, although not recommending joint modeling, noted
the interrelationship between defaults and prepayments. Fannie Mae
favored the use of a statistical model that would determine only total
terminations (prepayments plus defaults) in each of the two stress test
interest rate scenarios. Fannie Mae suggested that total defaults in
both scenarios be set at the levels that occurred in the benchmark loss
experience. Prepayments would be calculated by subtracting total
defaults from total terminations. Fannie Mae made no specific
recommendation about how conditional default rates might be determined
or how total defaults and prepayments should be distributed through the
stress period. Fannie Mae opined that the methodology it recommended
would be consistent with the 1992 Act and would provide a workable
framework for capturing the relationship between defaults and
prepayments. Fannie Mae also viewed this approach as consistent with
industry practice and asserted that it would be easier for the company
to manage to a capital standard based upon such an approach than it
would be to manage to one based upon a joint statistical model.
By contrast, Freddie Mac, while preferring a simpler approach to
default modeling, asserted that a joint statistical model of default
and prepayment rates would be preferable to total termination models in
the stress test context because: (1) unlike the total terminations
models, the joint model ensures that defaults and prepayments ``add
up'' to the total mortgage terminations; (2) total termination models
focus on interest rate movements under the assumption that default is a
small part of terminations under normal conditions, (an assumption
Freddie Mac found unwarranted in a stress test environment); and, (3)
standard termination models capture small effects such as seasonal
variation, which would unnecessarily complicate the stress test.
Freddie Mac also favored an empirically based statistical model of
mortgage performance over a stochastic simulation model like those used
in mortgage-backed security pricing. Freddie Mac stated that stochastic
models are not typically used by the industry for default and
prepayment modeling because borrower housing objectives are too complex
and heterogeneous to be described adequately with a single set of rules
simple enough to solve analytically.
Although Freddie Mac favored the use of a joint statistical model
over these other approaches, Freddie Mac did not recommend that OFHEO
use one in the stress test, asserting that OFHEO would have difficulty
using the data from the benchmark experience to estimate the model.
Freddie Mac also cited the need to model prepayments during the stress
period as a function of current coupons and interest rates. Freddie Mac
instead recommended estimating a statistical equation for prepayments
based on historical data from a distressed region to factor prepayments
into the stress test. Freddie Mac asserted that this approach would
allow implementation of the two interest-rate scenarios while tying
prepayment rates to the benchmark experience. Freddie Mac also
recommended using cumulative default rates from the benchmark
experience as the stress test default rates.
Freddie Mac raised other issues about joint models, claiming that
they are not ideal because: (1) they are complex; (2) they require
assumptions about both house price drift (average appreciation) and
volatility (variation in individual appreciation rates around the
average rate); (3) they require assumptions as to what constitutes
negative equity; and (4) they require other factors, such as loss of
employment to be modeled.
(ii) OFHEO's Response
OFHEO proposes to use joint statistical models in the stress test
for both single family and multifamily loans, agreeing with
recommendations of many commenters. Also, OFHEO found that total
termination models, such as those recommended by Fannie Mae, were not
adequate for the purposes of the proposed regulation. (See earlier
discussion under section III.A.5.a., Use of Conditional Default and
Prepayment Rates.) As explained in the ANPR, prepayments have a major
impact on cumulative and conditional rates of default, because every
loan that prepays is one less loan that could later default. However,
high levels of prepayment, which occur when interest rates decline, can
also result in increased conditional default rates in periods that
follow. This phenomenon, referred to as ``adverse selection'' or
``burnout,'' occurs because loans that do not prepay when interest
rates decline are often lower quality loans that do not qualify for
refinancing. Using a joint default/prepayment model allows the stress
test to reflect the impact of prepayments (and, therefore, of interest
rate changes) upon defaults.
The joint modeling approach is based on well-known and accepted
statistical methods that are widely applied in the mortgage performance
research. Researchers have found multivariate statistical models to be
necessary for this research, because the borrower's options to default
or prepay are interrelated. OFHEO believes that simpler approaches
(models or tabulations) that fail to account for this complexity would
not provide reasonable and appropriate projections of mortgage
performance during the stress period.
OFHEO addressed Freddie Mac's concern about the difficulty of
retaining a reasonable relationship to the benchmark loss experience in
a joint model by: (1) replicating certain benchmark economic factors--
specifically, house prices, rent growth rates and rental vacancy
rates--in the stress test; and (2) adjusting the underlying default and
severity equations used in the stress test to allow them to replicate
exactly the benchmark experience. Modeling the effects of differences
in starting coupons and interest rates from the benchmark loss
experience was possible, because OFHEO's database allowed the models to
be estimated based upon a broad and representative sample of historical
mortgage performance data. The statistical equations therefore yield
reasonable estimates that can be used to project mortgage prepayment
under many different circumstances, including stress test interest rate
scenarios.
Regarding the issue of model complexity, in OFHEO's view, the
proposed models strike the appropriate balance between accuracy and
simplicity. The stress test uses an approach based on well-known and
[[Page 18129]]
accepted statistical methods that are applied and accepted widely in
academic research and in industry practice. Further, OFHEO has
developed specifications for the default and prepayment models that
avoid unnecessary complexity. The prepayment model suggested by Freddie
Mac--using Freddie Mac projections from a statistical equation with ad
hoc adjustments based on mortgage coupon rates--is at least as complex,
but far less accurate.
As to house price appreciation and volatility, any model of
mortgage performance includes, explicitly or implicitly, assumptions
about these factors. OFHEO believes that the proposed stress test
includes a reasonable and appropriate methodology for updating house
prices throughout the stress period. (See section III.A.4.d., Property
Valuation.)
OFHEO does not agree with Freddie Mac that the need to use
assumptions about negative equity to estimate a joint model is a reason
not to use a joint model. Any statistical model of mortgage default
requires certain assumptions about how to measure negative equity in
order to predict defaults. Although expected equity values cannot be
assigned to individual borrowers to determine a precise LTV for each
loan, using probabilities of negative equity provides substantial
information about the negative equity position of individual borrowers.
The probability of negative equity is a function of the current loan
balance and the probability that individual house prices are below that
balance. It is especially valuable when modeling the default potential
from groups of loans, as is the case in the proposed stress test. By
applying estimates of house price drift and volatility obtained from
independent estimates based on the OFHEO House Price Index, the
distributions of individual housing values relative to the value at
mortgage origination are determined. This approach eliminates the
measurement difficulties associated with calculating individual
borrower equity at the loan level.
The concern that developing a statistical model for the stress test
would require modeling the effects of unemployment on prepayment rates
does not raise an issue, because OFHEO does not propose to use
unemployment as an explanatory variable in the stress test. In general,
OFHEO has limited the explanatory variables in the stress test to those
that define different loan characteristics or product types are
required to meet statutory requirements. As explained above in section
III.A.2., Overview of Mortgage Performance, OFHEO has avoided
variables, such as unemployment, that require assumptions about stress
period economic conditions that are not specified in the 1992 Act. (See
section III.A.5.e., Choice of Explanatory Variables for Default and
Prepayment).
d. Choice of a Statistical Method for a Joint Model of Default and
Prepayment
(i) ANPR Comments
The ANPR sought comment on the appropriate statistical method to
use for a joint model of default and prepayment. None of the ANPR
comments provided an express recommendation of a model, but NAR
supported a multivariate model and suggested that the proportional
hazard model developed by John Quigley and Robert Van Order in 1992
would provide a good starting point. Other commenters, such as Freddie
Mac and ACB, emphasized that any joint model must be robust and able to
yield reasonable results under many different scenarios.
(ii) OFHEO Response
OFHEO agrees with the NAR comment that proportional hazard models
provide a good starting point. These models measure conditional rates
of default and prepayment. The stress test utilizes a similar approach,
the logit model, which is more appropriate for large data sets. OFHEO
also agrees with Freddie Mac and ACB that a joint model should be
robust and able to yield reasonable results under many different
scenarios. As explained more fully in the Technical Supplement, OFHEO
has evaluated its proposed models to ensure that they yield reasonable
results under many different scenarios, use widely accepted techniques,
and are otherwise appropriate for OFHEO's purposes.
OFHEO is proposing statistical models for single family mortgages
that were estimated using multinomial logit specifications for
quarterly conditional probabilities of default and prepayment. The
multifamily model was estimated similarly, although it is based upon
annual, rather than quarterly, conditional probabilities of default and
prepayment, as described more fully in the discussion of the
multifamily default/prepayment issues, below. There are several
advantages to using the multinomial logit specification. First, it
guarantees that the estimated and projected probabilities of default
and prepayment always lie between 0 and 100 percent. Second, one can
estimate weights for the impact of specific explanatory variables on
the probabilities of default and prepayment separately. Third, it is
possible to specify different lists of explanatory variables for each
type of event. Fourth, the model automatically accounts for the impact
of differences in the estimated probability of default on prepayment
and vice versa. Finally, estimation routines for multinomial logit
models are readily available in a large number of commercially
available statistical software packages.
e. Choice of Explanatory Variables for Default and Prepayment
In the ANPR, OFHEO requested comment on the appropriate explanatory
variables to use in statistical models of default and prepayment. OFHEO
asked specifically about how to account for the effects of house
prices, interest rates, and other economic factors, and whether to
include measures of mortgage age and mortgage value as explanatory
variables. OFHEO also asked about empirical and theoretical approaches
to estimation of multifamily credit risk, and several respondents
addressed the issue of explanatory variables in responding to that
question.\92\ Because there are some differences between the
explanatory variables for single family and multifamily models, the
comments on explanatory variables are discussed separately for the two
models. Some comments related to specific explanatory variables are
discussed below in connection with the discussion of the particular
variable.
---------------------------------------------------------------------------
\92\ No commenters provided suggestions on how to actually model
multifamily mortgage defaults and prepayments.
---------------------------------------------------------------------------
(i) Comments on Explanatory Variables for Single Family Modeling
Freddie Mac suggested that using mortgage product, property type,
occupancy status and current LTV as explanatory variables would explain
a significant portion of the differences in default rates without
venturing into more complex relationships that might prove unreliable
for purposes of the stress test. Freddie Mac recommended caution in the
consideration of mortgage age as an explanatory variable, noting that
while age may be a valuable proxy for unmeasurable determinants of
default, it should not take on such importance that mortgage age
patterns dominate the capital requirements. In contrast, Freddie Mac
did recommend that OFHEO include a measure of the mortgage premium
value (reflected by the difference between the interest rate on a given
mortgage and the current market interest rate for a similar loan) in
[[Page 18130]]
its modeling efforts, as an adjunct to borrower equity. Freddie Mac
cited its own research showing that borrower default choices do respond
to differences between the mortgage coupon rates and current market
rates of interest.
World Savings stated that OFHEO should be cautious about including
unemployment rates as an explanatory variable in any statistical model
of mortgage performance, because the statutory stress test takes a
regional experience and uses it to imply a national recession. World
Savings reasoned that, in a regional recession, homeowners who lose
their jobs might find employment elsewhere but retain their homes. They
may rent their homes until such time as house prices again rise enough
to permit them to sell their properties without incurring a loss.
However, in a national recession, such opportunities would not be
available and the dynamics of default could be much different.
MRAC recommended using the following variables: current LTV, length
of residence, mortgage term and type, loan purpose, occupancy status,
primary home status, relocation loan status, consumer credit
information, and mortgage premium value. Recognizing that length of
residence is not always available to researchers, MRAC suggested that
mortgage age could be used instead. The MBA recommended including
measures of borrower equity, mortgage premium value, and product type
differences in a statistical model. Standard and Poor's asserted that
mortgage age is a very important explanatory factor, noting that 80
percent of all defaults occur by the seventh year of a mortgage pool.
The VA asserted that borrower equity is the most important
determinant of default and prepayment rates and recommended that OFHEO
think of explanatory variables in two categories: those that indicate
the borrower's ability to pay, and those that indicate the borrower's
ability to sell the property. The former category could include such
things as job loss, divorce, necessary relocation, and hazard loss
(e.g., uninsured fire or water damage to the home). The latter category
could include the borrower's equity position and ability to complete a
property sale quickly. The VA also mentioned that its own statistical
model of default and prepayment rates includes regional unemployment,
house sale activity measures, and a house-purchase-affordability index.
NAR recommended that OFHEO include a factor for mortgage age, but
not for the mortgage premium value. While NAR accepted the theoretical
justification for including mortgage value in a statistical model, it
did not find its influence on defaults to be statistically significant
in its own modeling efforts. NAR also mentioned a factor not discussed
by other commenters--the relative size of each loan. NAR commented that
the influence of house price appreciation on default depends on whether
the loan has a high or low balance, and that OFHEO should carefully
analyze this issue in the context of Enterprise experience. In addition
to these comments, NAR also provided, without further explanation, a
list of all the variables it believes should be included in a
statistical model of default and prepayments. Listed were: origination
LTV, ratio of the mortgage coupon rate to the current market rate for
home mortgages, current LTV, loan size, presence of credit enhancement
(e.g., private mortgage insurance), house price dispersion, transaction
costs, the burden on household cash flow of servicing the mortgage,
origination year of the mortgage, policy year (age) of the mortgage,
mortgage premium value (for prepayment only), region of the country,
unemployment rate, inflation, regional household mobility rate,
mortgage product characteristics, and net borrower equity in the home.
(ii) Comments on Explanatory Variables for Multifamily Modeling
OFHEO received fewer responses to its ANPR questions on approaches
to multifamily modeling than it did to questions related to single
family mortgage performance modeling. The import of these comments was
to direct OFHEO to look at property cash flows as the primary influence
on defaults. Freddie Mac emphasized that cash flow after mortgage debt
service, as measured by the debt coverage ratio (DCR) is important, as
are property equity and balloon terms. It also mentioned the need to
measure multifamily market conditions directly, rather than relying
upon single family house price appreciation to update explanatory
variables over time. Freddie Mac further indicated that OFHEO needs to
take into account significant factors that affected multifamily default
rates during the 1980s, such as tax law changes, but should not include
in the stress test the effect of any speculative political factors,
such as potential legislative actions.
Standard and Poor's also suggested that DCR should be the focal
point for multifamily mortgage default risk, but added that the quality
of the real estate securing mortgages is also considered in the S&P
credit analysis. ACB recommended accounting for the changing cash flow
position of the mortgaged property (i.e., using the DCR), rather than
relying solely on net income, and including factors for tax laws and
depreciation allowances. It also commented that, while data is not
available to consider these additional variables, the underlying
determinants of multifamily defaults are factors that lead to problems
in tenant rental payments: unemployment, reduced hours of work, and
reduced income. HUD suggested considering the corporate bankruptcy
literature when deciding how to model multifamily defaults. This
literature emphasizes changes in the cash flow position of multifamily
properties. HUD also commented that OFHEO should treat balloon payoffs
differently than normal, early prepayments.
(iii) General Approach
Models of mortgage performance are models of borrower behavior--of
individual borrowers' decisions whether to continue making monthly
mortgage payments, to prepay, or to default. Each month, every borrower
must choose among these three options. Because mortgage performance
models are an attempt to predict how borrowers will choose to exercise
these options, financial options theory provides the most widely
accepted conceptual framework to link these borrower choices to
differences in the underlying loan characteristics and economic
conditions.\93\
---------------------------------------------------------------------------
\93\ This conceptual framework is the basis for nearly all
mortgage performance research. It applies to all of the mortgage
performance models referenced in the ANPR (See 60 FR 7470-7471, Feb.
8, 1995, footnotes 11 and 13). Other references can be found in the
Technical Supplement to this regulation. Financial options theory
treats a mortgage like a bond issued by the borrower with embedded
financial options to default or prepay, which borrowers will
exercise when it is in their financial interest to do. From the
lender or mortgage investor's perspective, this conceptual framework
is sometimes referred to as ``contingent claim analysis.'' The
mortgage investor, as bondholder, has a claim to a cash flow
(mortgage payments), the value of which is contingent upon the value
of the options to the borrower and the actions of the borrower with
respect to the mortgage property (e.g., property maintenance). The
choice to pay off (prepay) a mortgage is likened to a ``call''
option, where the borrower effectively buys back the mortgage from
the lender at the book (face) value. The choice to default is seen
as a ``put'' option, where the borrower sells the mortgage back to
the lender at the current market value of the collateral property.
The choice of an options-based model is consistent with the apparent
underlying assumption of the preponderance of ANPR comments, which
generally relate to how to account for factors that affect the
exercise of these options.
---------------------------------------------------------------------------
In the options theory framework, the most important variables are
borrower equity and interest rates. When equity is
[[Page 18131]]
negative, that is, the property value is less than the outstanding
mortgage balance, the default (put) option is said to be ``in the
money.'' That term is used to mean that, theoretically, the borrower
might find it financially advantageous to default in order to eliminate
the negative equity position in the mortgage.\94\ When equity is
negative, maintaining the mortgage through regular monthly payments
leaves the borrower paying more for the property than it is worth.
Under such conditions, default becomes an economically rational option
for many borrowers, particularly those who may be undergoing other
financial stresses, such as unemployment, divorce, health problems,
etc.
---------------------------------------------------------------------------
\94\ Negative equity is only one factor that influences the
borrower's decision. Borrowers are usually personally liable on the
note, which means that default could have numerous negative
consequences beyond losing the property in foreclosure. For this
reason, the model recognizes that negative equity does not cause a
default, but simply makes it more likely.
---------------------------------------------------------------------------
In an options-based model, interest rate changes create positive or
negative value in the mortgage itself. This value is referred to in the
ANPR as ``mortgage value.'' It is also sometimes referred to as the
mortgage premium value. That is, the current mortgage has a ``premium''
or positive value to the borrower--it is worth holding on to--if the
coupon interest rate is below current market rates. That mortgage value
is reduced if current market rates are below the coupon rate. If a
borrower is in a position of negative property equity due to declines
in local house prices, but has a below market rate mortgage, the
mortgage premium value reduces incentives to default. On the other
hand, an above market rate mortgage could, in theory, increase the
incentive to default for the same borrower.
The mortgage premium value is inversely related to the value of the
prepayment (call) option. When current market rates are below mortgage
coupon, the call option is ``in the money,'' and its value is high.
When the mortgage rate is below market, the call option is ``out of the
money,'' and its value is low. Borrower equity also plays a part in
prepayment determination; generally, it must be a certain positive
amount before lenders will offer refinance opportunities. It must also
meet a positive threshold before a property can be sold without the
borrower incurring out-of-pocket expenses. However, as long as minimum
equity thresholds are met, the higher the mortgage coupon rate is above
the market rate, the greater is the incentive for a borrower to
exercise the prepayment option by paying off the existing mortgage from
the lender with the proceeds of a new loan.\95\
---------------------------------------------------------------------------
\95\ It is also possible that borrowers exercise the prepayment
option with personal equity, liquidating other assets to pay off the
mortgage even if property equity is negative. Borrowers may also
turn to alternate lenders, who offer loans with LTVs higher than
those usually purchased by the Enterprises, for refinancing
opportunities when borrowers have little or no positive property
equity.
---------------------------------------------------------------------------
Although property equity and interest rates are the predominant
variables of relevance in an options approach to mortgage termination
modeling, many other factors affect borrower decisions to exercise a
default or prepayment option.\96\ For single family mortgages, some of
these factors are: (1) the potential for lender deficiency judgments,
which reduce borrowers' ability to force lenders to absorb the negative
property equity through defaulting; (2) borrowers' desire to maintain
access to credit at preferential rates, which will also make them more
hesitant to default; (3) moving costs, which reduce the value of the
default option; (4) forced mobility due to job loss (or relocation) or
family disruption, causing default or prepayment when it would not
otherwise be financially advantageous to terminate the mortgage; (5)
expected future mobility, which reduces tendencies to prepay in the
present when that option is otherwise ``in the money''; and (6) the up-
front expenses involved in prepayment, which require that interest
rates fall by a certain amount before it is really advantageous to
prepay. For multifamily mortgages, the additional factors that affect
the borrower's decision to exercise an option to default or prepay are:
(1) property cash flow and the ability to service the mortgage; (2) the
value of depreciation write-offs in reducing tax burdens; (3)
prepayment penalties, which reduce the value of refinancing in the
early years of a loan; and (4) balloon terms, which generally require a
loan to be refinanced at maturity. Balloon term considerations are more
important for multifamily than for single family mortgages because
balloons are the predominant instrument type in the conventional,
multifamily mortgage market.
---------------------------------------------------------------------------
\96\ Empirical studies have shown that mortgage borrowers are
not ``ruthless'' in their exercise of these options. First, just
being ``in the money'' at a point in time does not mean that an
optimal ``strike price'' has been reached, where the option value is
maximized. Second, there are many other factors that affect both
option value and whether borrowers will default or prepay their
mortgages.
---------------------------------------------------------------------------
In choosing which variables to include in estimating the
statistical models used in the stress test, OFHEO considered financial
options theory, ANPR comments, data availability, the need for
simplicity in model design, and the need to meet multiple statutory
objectives while implementing a credit stress test based on the
benchmark loss experience. In selecting explanatory variables to use in
running the stress test, OFHEO considered whether they were necessary
to reflect the differences in loan characteristics and interest rate
environments as required by the 1992 Act. Some variables were used to
estimate the statistical models, but they did not meet the criteria for
inclusion in the stress test itself.\97\ They are represented by
simplifying assumptions in the stress test so that their values do not
vary across loans or time. All variables used to estimate the models
and any other variables suggested by commenters are discussed below.
The variables common to both single family and multifamily analysis are
discussed first, followed by a discussion of variables unique to each.
---------------------------------------------------------------------------
\97\ Any variable that is included as an explanatory variable in
the stress test is also used to estimate the model.
---------------------------------------------------------------------------
(iv) Common Single and Multifamily Variables
(a) Measures of Borrower Equity
The actual variable used in the proposed stress test to capture
borrower equity positions is the probability of negative equity--the
probability that the value of a mortgage will be larger than the value
of the property securing it, so that the default (put) option is ``in
the money.'' Calculation of this explanatory variable uses the measures
of property value described in section III. A. 4. d., Property
Valuation, along with original loan amortization schedules.\98\
Measuring the probability of negative equity is appropriate because the
actual appreciation rates of individual properties are unknown and
because such a measure gives the best representation of the percentage
of loans in any given pool or portfolio that are at risk of default.
The probability of negative equity is also included in prepayment
equations, because negative equity may prevent prepayment by making it
difficult to refinance. This variable, therefore, has opposite effects
on default and prepayment rates. Increases in the probability of
negative equity mean that fewer loans in the pool qualify for
refinancing, which decreases prepayment rates. At the same time,
borrowers who are forced to relocate or
[[Page 18132]]
who experience a loss of income may have difficulty prepaying, making
the default option a more likely borrower strategy.
---------------------------------------------------------------------------
\98\ In the estimation of single family default and prepayment
equations, and in the stress test simulation of default and
prepayment rates, balloon loans are amortized over their original
rather than amortization terms. In the final rule OFHEO intends to
substitute amortization term for original term in the calculations
for balloon loans.
---------------------------------------------------------------------------
For multifamily loans, the stress test uses a variable capturing
the joint probability of negative equity and negative cash flow to
predict default. As highlighted by the ANPR commenters, cash flow may
be more important than equity for multifamily default. Although
negative equity is a necessary condition for the default option to be
``in the money,'' it is not a sufficient condition for default. Default
will maximize wealth only if cash flows are also negative. When the
equity is negative, but cash flows are positive, default is not
rational because the borrower would give up positive income. Because
both negative equity and negative cash flow are required for default to
occur, the primary variable proposed to explain multifamily default is
the joint probability that a property has both negative equity and
negative cash flow.
Additional consideration is given to the equity position of
borrowers with balloon loans when those loans mature. At the balloon
maturity point, when borrowers must pay off and find new financing,
weak property financials can lead to even higher default rates than
might occur earlier in the life of the loans. The multifamily model,
therefore, gives additional weight to the joint probability variable in
the balloon maturity year to reflect the increased risk that a borrower
will not qualify for a new mortgage.\99\
---------------------------------------------------------------------------
\99\ OFHEO does not propose a similar treatment of single family
balloon mortgages at this time, because they are not substantial
portions of single family loan portfolios of the Enterprises, their
balloon point refinance qualification standards are not as stringent
as those for multifamily loans, and the Enterprises readily help
single family borrowers to refinance balloon mortgages.
---------------------------------------------------------------------------
Multifamily balloon loan payoff is also a function of the financial
characteristics of the underlying property, because loans must meet
equity and cash flow standards before new financing can be secured. To
capture the impact of equity and cash flow on the ability of a borrower
to refinance a multifamily loan at the balloon point, the stress test
uses a variable that measures the joint probabilities that both
property equity and cash flow are at sufficiently high levels to
qualify for refinancing.
(b) Mortgage Premium Value
OFHEO posed a question in the ANPR about use of the mortgage value
(mortgage premium value)--the financial value of an above or below
market rate mortgage coupon--as an explanatory variable in default
equations. The mortgage premium value is a measure of the value of the
prepayment option to the borrower, that is, the value of prepayment
before accounting the transaction costs of prepayment. It is,
therefore, an important variable used by all the models to explain
prepayment behavior. At issue is whether this factor should also be
used to help explain default behavior.
ANPR commenters had differing views on this issue. Those suggesting
that it should be used were Freddie Mac and VA. Two other commenters,
NAR and ACB, were supportive in theory, but were not confident that a
statistically valid relationship to default rates could be found, at
least for single family mortgages. MRAC included the difference between
the mortgage coupon rate and current market interest rates (a proxy for
mortgage premium value) in its list of explanatory variables for a
default/prepayment model. This is a proxy for the mortgage premium
value.
As explained earlier, options theory suggests that increases in the
value of the prepayment option (resulting from lower interest rates)
should increase both prepayment and default rates because the current
mortgage becomes expensive compared to alternatives. Prepayments
increase because refinancing becomes attractive. Default rates increase
for borrowers who already have negative property equity because some
such borrowers relieve themselves of both the negative property equity
and the expensive mortgage by defaulting and then renting, or by taking
out a new mortgage to purchase another property. Conversely, increases
in market interest rates increase the value of holding on to an
existing mortgage, and thus may decrease default rates as well as
prepayments.
While recognizing that there is a theoretical basis to include a
mortgage premium value variable in the default equations, OFHEO
proposes, nevertheless, to limit its use to prepayment equations. The
influence of interest rate changes on mortgage defaults is captured
adequately in single family default equations by a ``burnout''
variable, which measures the instances when borrowers have not taken
advantage of previous refinancing opportunities. This variable is
explained in a later discussion under section III.A.5.e., Choice of
Explanatory Variables for Default and Prepayment. A burnout variable is
not included in the multifamily equations, because prepayments are
severely limited by prepayment restrictions.
For prepayment equations, the actual variable used to capture the
prepayment option value is a relative spread variable: the difference
between the current mortgage coupon rate and the current market
interest rate, as a percentage of the current mortgage coupon rate.
This variable has been shown to provide an approximation of the
mortgage premium value.\100\
---------------------------------------------------------------------------
\100\ This approximation of the mortgage premium value was
introduced by Y. Deng, J. M. Quigley, and R. Van Order, (1996)
``Mortgage Default And Low Downpayment Loans: The Costs Of Public
Subsidy,'' Journal of Regional Science and Urban Economics 26(3-4),
263-285.
---------------------------------------------------------------------------
For multifamily mortgages, this relative spread variable is not
included in the default equations, because the interest rate effect on
default rates is reflected adequately in the joint probability
variable. Declines in interest rates increase the present value of
after-debt income stream generated by the property, and thus its market
value, all else equal. Consequently, multifamily property values
generally rise when interest rates fall.\101\ Thus, a relative spread
variable is not included for multifamily defaults.
---------------------------------------------------------------------------
\101\ While market interest rates do have some effect on prices
of single family homes, the effect is not as direct as it is for
multifamily and other investment properties.
---------------------------------------------------------------------------
(c) Mortgage Age
OFHEO proposes to include mortgage age as an explanatory variable
in its single family and multifamily models, as recommended in the ANPR
comments. OFHEO found that conditional probabilities of default and
prepayment of Enterprise loans exhibit characteristic age profiles that
increase during the first years following origination, peak sometime
between the fourth and seventh years, and decline thereafter.
Because the benchmark loss experience was based entirely upon newly
originated loans, an adjustment is necessary to account for the fact
that at any point in time Enterprise single family portfolios consist
of loans with varying ages. Adding mortgage age as an explanatory
variable provides such an adjustment by allowing conditional default
and prepayment probabilities to vary during the stress period in ways
that historical profiles indicate are appropriate for loans of each
age. Although Freddie Mac raised a concern that mortgage age might have
too large an effect in the stress test, OFHEO research indicates that
this is not the case. Although mortgage age is an important variable in
the models, it does not diminish the impact of other, more
[[Page 18133]]
direct risk factors included in the stress test.\102\
---------------------------------------------------------------------------
\102\ Mortgage age combines with the constant term in the
statistical default and prepayment equations to create what can be
called ``baseline'' rates of default and prepayment: the time series
of rates that would occur if all other influences were absent. Once
variables representing those other influences are added to the
equations, the actual patterns of default and prepayment rates can
vary greatly from the baseline paths.
---------------------------------------------------------------------------
(v) Additional Explanatory Variables Used in the Single Family Model
The following discussion addresses additional explanatory variables
that are used only in the single family model. A list of additional
explanatory variables for the multifamily model is provided after this
discussion of single family variables. The variables discussed below
help to complete or modify the basic option valuation for single family
mortgages. The original LTV ratio helps to account for differences in
default and prepayment rates due to borrower financial status.
Occupancy status accounts for differences between single family owner-
occupiers and investor-owners. Product-type factors adjust for
differences that might be due to the unique risk characteristics of
those products and the borrowers who use them. The yield curve slope
accounts for different incentives to refinance between fixed-and
adjustable-rate products. Some of the variables discussed below are
used in statistical estimation of the models, but are represented by
simplifying assumptions in the stress test.
(a) Original LTV Ratio
Original LTV ratio is used in the stress test as a proxy for a
number of factors related to the financial status of single family
borrowers that are recognized widely as influencing the propensity of
borrowers to default. Among these factors, which were mentioned by ANPR
comments, are borrower income, net worth, and debt burdens. Information
about these factors is not available for most of the loans in OFHEO's
database. A variable that is available as a proxy for relative
financial status of borrowers is the original LTV ratio.\103\ Both
Freddie Mac and NAR recommended use of this variable. By making low
down payments, high LTV borrowers signal that they are more likely to
have few economic resources to finance the transaction costs of
prepayment, or to endure spells of unemployment or other ``trigger''
events that might cause them to exercise their option to default. Also,
high LTV borrowers demonstrate a willingness to ``leverage'' the
financing of the home purchase, which may mean that they are more
likely to exercise their default option when it is in the money. For
these reasons, OFHEO found that original LTV is an important risk
characteristic of mortgages, which OFHEO proposes to use both in
estimating the single family model and in running the stress test.
---------------------------------------------------------------------------
\103\ Although credit scores could be a good indicator of the
financial status of borrowers, as discussed below under section III.
A. 5. e. vi. f., Credit Scores, their usefulness for developing and
implementing a default/prepayment model in the stress test is
limited because credit scoring is a fairly recent development in the
mortgage industry.
---------------------------------------------------------------------------
(b) Occupancy Status
Historically, single family loans to owners who live in the
collateral property have exhibited different performance than similar
loans made to investors who rent the property. Difference in occupancy
status is one of the loan characteristics that the 1992 Act
specifically requires that OFHEO take into account in the stress test.
It is also a distinction often made by the mortgage industry, because
of a clear difference in the risks of borrower default or prepayment.
Owner occupants are less likely than investors to exercise the default
option because of the direct benefits occupants receive from the
consumption of housing services. Also, owner occupants are more likely
to prepay for non-financial reasons, such as residential mobility, than
are investors.
The statistical equations used in the stress test were estimated
with an investor loan indicator variable that captures the differential
default and prepayment risk of these mortgages. However, to capture the
differential risk of investor loans in the proposed stress test, OFHEO
makes a simplifying assumption that investor loans are spread equally
across all loan groups, according to their percentage in the overall
Enterprise book of business, rather than creating separate loan groups
for investor mortgages. For example, if investor loans are four percent
of all loans for a particular Enterprise in a particular starting
quarter for the stress test, then four percent of the loans in each
aggregated loan group are presumed to be investor loans for purposes of
running the stress test. The statistically derived investor-loan
weighting factor (statistical coefficient) in each default and
prepayment equation is then applied to the four percent figure to
arrive at the differential investor loan risk for every loan group.
Because investor loans are a small percentage of Enterprise single
family portfolios and are heavily concentrated in the 70 to 80 percent
LTV category, OFHEO's simplifying approach has no significant impact on
loss rates.\104\ The exact algorithms used in the proposed stress test
to capture investor loan risk are detailed in section 3.5.2.3.2.5.,
Occupancy Status (OS), of the Regulation Appendix.
---------------------------------------------------------------------------
\104\ Loans on owner-occupied properties in the Enterprise
portfolios also have a central LTV range of 70-80 percent. Thus,
attributing some investor loans to higher LTV categories and some to
lower categories, by assuming they have the same overall LTV
distribution as do owner-occupied loans, has offsetting effects on
predicted credit risk.
---------------------------------------------------------------------------
(c) Product Type
The 1992 Act expressly requires OFHEO to take differences in
mortgage product type into account. In addition, because the benchmark
loss experience was identified using the 30-year fixed-rate mortgage,
it is necessary to reasonably relate the default experience of other
types of mortgage products to the benchmark. Most commenters suggested
some type of multiplier approach for other single family mortgage types
that would measure the risk of these products in proportion to the risk
of the benchmark loan type. OFHEO's proposed approach is broadly
consistent with the thrust of these comments. Because comments received
by OFHEO focused particularly on relating various mortgage product
types to the benchmark experience, these comments are discussed later
under section III.A.7.b., Relating Other Single Family Products to the
Benchmark. This section discusses the way in which mortgage product
type differences are handled in the single family mortgage performance
model.
The stress test uses two primary sets of statistically estimated
single family default/prepayment equations, one for fixed-rate and one
for adjustable-rate mortgages. A third set of equations, which may be
thought of as modified fixed-rate equations, is used to project the
performance of less prevalent single family mortgage types relative to
the performance of 30-year FRMs. This final set of equations includes
as explanatory variables unique product-type indicators for 15-year
fixed-rate mortgages, 20-year fixed-rate mortgages, balloon mortgages,
FHA/VA-insured mortgages, and second liens. Description of these
specific product-type variables and their derivations are included in
section 3.5.2.3.2.8., Product Type Adjustment Factors of the Regulation
Appendix and section IV.B.5.j., Product Type Indicators, of the
Technical Supplement. Product type indicators allow estimation of
multiplier-like effects using all available historical data, and they
assure that measured differences in product-type
[[Page 18134]]
risk are consistent with the stress test environment. All products with
variable payments over time are included as adjustable-rate mortgages.
Other non-standard mortgage types, such as reverse mortgages and bi-
weekly mortgages, are included with their fixed-rate counterparts with
similar mortgage contract terms (length of mortgage in years).
As explained in section III.A.7.b., Relating Other Single Family
Products to the Benchmark, some commenters were justifiably concerned
that applying several product type multiples to a single loan would
have an inappropriate compounding effect on default rates. OFHEO
addressed these concerns in two ways. First, the multipliers were
estimated in a multivariate statistical analysis within the default and
prepayment probability equations, rather than applying fixed
multipliers to estimated default rates for 30-year fixed-rate loans.
This approach provides adjustment factors that are most consistent with
broad historical experience and with the other risk factors in the
model. By controlling for other explanatory variables, only the
residual effects of the differences in product type are captured by
these product-type adjustment-factor multipliers, which limits the size
of their effects. Second, the models include all other explanatory
variables as categorical variables (indicators of value-range
categories), instead of as continuous measures of variable values.
Using categorical variables helps control for unreasonable compounding
risks, by preventing the combination of low house-price growth and
sustained adverse interest-rate movements in the stress test to cause
default rates to rise to unrealistic levels. For example, the stress
test gives the same default weight to all probability of negative
equity values above 35 percent, which effectively caps the influence of
this variable in the stress test.\105\
---------------------------------------------------------------------------
\105\ The number of loans in the historic sample used to
estimate the statistical model of default and prepayment rates gets
very small as the value of the probability of negative equity rises
much above 35 percent. OFHEO therefore does not believe that there
is valid information on default risk that could be gained by
allowing for categories of probability of negative equity above, for
example, 50 percent.
---------------------------------------------------------------------------
(d) Yield Curve Slope
The slope of the Treasury yield curve is included as an explanatory
variable in the prepayment equations. Both the choice between ARM and
FRM loans and the timing of refinancing are influenced by expectations
about future interest rates and differences in short-term and long-term
borrowing rates associated with the slope of the Treasury yield curve.
The slope of the Treasury yield curve is measured in the proposed
stress test by the ratio of the ten-year CMT to the one-year CMT. A
high value for the slope of the yield curve indicates that short-term
rates are low relative to long-term rates. A high value, therefore,
reduces the likelihood that ARM borrowers will refinance into fixed-
rate mortgages, and increases the likelihood that fixed-rate borrowers
will refinance into ARMs to take advantage of the more attractive
interest rates.
(e) Burnout
For single family mortgages, the proposed stress test uses the
variable burnout to capture the effect of the inability of borrowers to
refinance their mortgages due to equity or other credit constraints.
Burnout is the adverse selection that occurs when borrowers retain
their mortgages during periods when there are clear financial benefits
to refinancing. In this context, adverse selection is reflected in the
lower average credit quality of mortgages remaining in a pool after a
significant refinancing opportunity, compared to the overall quality of
the mortgages in the original, larger pool. Adverse selection occurs
because borrowers and properties with higher credit quality refinance
in higher proportions than do those with lower credit quality. The
remaining mortgages, therefore, will experience higher conditional
default rates. Accounting for this change in the underlying quality of
a mortgage pool is preferable to using only a prepayment-option-value
variable in predicting defaults, principally because its effect
continues unchanged over time. The burnout variable in the stress test
indicates whether, over the previous eight quarters of mortgage life,
there have been at least two quarters with significant refinance
opportunities, as defined by a two percentage point difference between
the mortgage coupon rate and the market interest rate on fixed-rate
mortgages.
For similar reasons, burnout is also included as an explanatory
variable in single family prepayment equations, although its effect is
in the opposite direction to that in the default equations. As
discussed in the ANPR, burnout suggests that prepayment rates will be
less responsive to interest rate changes after a pool of mortgages has
already undergone a significant period of refinance opportunities.
(vi) Single Family Variables Not Used in Running the Stress Test
Addressed below are several variables suggested by ANPR commenters
that either are not used in the single family default/prepayment model,
or were included in the statistical estimations but are represented by
fixed or constant values when the stress test is run. In general, to
estimate the model, OFHEO used variables that had significant
independent effects on default and prepayment rates. However, OFHEO
does not propose to use all of these variables in running the stress
test. Some variables are not used in the stress test because they would
diminish the role of the benchmark loss experience in determining
stress test credit risk. Others were not needed to reflect statutory
requirements to distinguish among loan types and characteristics, or
between the effects of the up-rate and down-rate scenarios. Allowing
such variables to vary in value in running the stress test would create
credit-risk dimensions that are unnecessary and not contemplated by the
statute.
(a) Relative Loan Size
Relative loan size \106\ is the ratio of the original loan amount
to the average-sized loan purchased by the Enterprises in the same
State and in the same origination year. This variable was included when
estimating the statistical model to isolate differences in the
performance of loans of above and below average size, but is not used
in the stress test.
---------------------------------------------------------------------------
\106\ Relative loan size should be distinguished from the actual
original and current dollar balances of the loans, which are
included elsewhere in the stress test.
---------------------------------------------------------------------------
As suggested by NAR, OFHEO explored the different default
propensities of loans with high and low balances using Enterprise data.
OFHEO's use of a relative loan size variable in the statistical
estimations of the single family model demonstrated that relatively
larger loans tend to have higher prepayment speeds, but differences in
default rates by loan size were small and inconsistent. OFHEO
interprets the faster prepayment speeds of relatively large loans as
reflective of the higher dollar value of the prepayment option on these
loans. Households with relatively large loans may also have higher
overall debt burdens and be more responsive to opportunities to
refinance debt so as to lower payment burdens.
The stress test does not use relative loan size as a variable,
because it is not needed to reflect statutorily required distinctions,
and including it as a variable would have necessitated a sevenfold
increase in the number of loan group records in the stress test. OFHEO
believed that the benefit
[[Page 18135]]
derived did not justify the additional complication of the stress test
that would result. As a result, all loans are put into the ``average''
size category for this variable when running the stress test.\107\
---------------------------------------------------------------------------
\107\ This value is part of the fixed-factor terms reported in
section 3.5.2.3.3., Combining Explanatory Variables and Weights of
the Regulation Appendix for each default and prepayment equation.
Relative loan size is discussed in section B.5.i., Relative Loan
Size of the Technical Supplement.
---------------------------------------------------------------------------
(b) Season of the Year
The season (quarter) of the calendar year was included when
estimating the statistical model to account for the potential impact of
weather, school schedules, and seasonal employment patterns on
residential mobility and default and prepayment. In order to avoid
seasonal variation in the quarterly risk-based capital requirements
when the model is applied in the proposed stress test, an average of
the season of the year effects is used. Because of the actual
statistical technique used to estimate the equations, this average
effect is obtained by excluding the season-of-year variable from the
stress test default and prepayment equation.\108\
---------------------------------------------------------------------------
\108\ Seasonal variation is discussed in section B.5.g., Season
of the Year, of the Technical Supplement.
---------------------------------------------------------------------------
Use of seasonal variation was mentioned by Freddie Mac as a
weakness of the termination models used by investment banks to value
mortgage backed security pools. OFHEO agrees with Freddie Mac that such
seasonal variation would complicate the stress test, by creating
quarterly volatility in loss rates, with no particular safety and
soundness benefit.
(c) Origination Year
Freddie Mac and NAR recommended including origination year as a
variable. This approach would capture differences in the performance of
specific mortgage origination cohorts due to excluded factors such as
regional income growth and unemployment, or changes in mortgage
underwriting standards over time. OFHEO considered using this variable
but found that origination year is not an inherent risk factor, is not
needed to reflect the types of distinction required by the 1992 Act,
and is incompatible with the requirement to relate stress test losses
to the benchmark loss experience. The last point is most important. The
benchmark loss experience captures loans with the worst origination
year and the worst credit risk profile. Assigning to loans originated
in a given year a unique underlying credit profile, which may be
different from the benchmark credit profile, would remove an important
element of the link between stress test losses and the benchmark loss
experience. In addition, varying inherent credit risk by loan
origination year would require speculative assumptions about loan
quality for more recent origination years for which no credit-risk
track record has yet been established.
By not including origination year as an explanatory variable, the
statistical equations capture average origination-year profiles of
default and prepayment. As discussed later under in section III.A.7.,
Relating Losses to the Benchmark Loss Experience, these profiles are
adjusted further to reasonably relate starting loan portfolios to the
benchmark loss experience. If the stress test were to allow for
origination year differences when estimating the statistical equations,
it would be necessary to assign the benchmark origination year effect
to all loans in the stress test to preserve a reasonable relation to
the benchmark loss experience. This approach would complicate the
stress test without changing the results that are obtained using the
proposed approach.
(d) Unemployment
Unemployment rates were listed by some commenters as a possible
explanatory variable. For numerous reasons, OFHEO does not propose to
include unemployment as a variable either in running the stress test or
in estimating the statistical model. OFHEO does not propose to include
unemployment rates as an explanatory variable in the stress test,
primarily because it is not a loan characteristic, but a macro-economic
variable, and it is not one of the economic variables specified in the
1992 Act. In any event, the effect of economic-condition variables not
specified in the statute, such as unemployment, are captured in the
stress test by relating the stress test to the actual benchmark loss
experience, because the appropriate values are inherent in that
experience. Thus, reasonably relating the stress test to the benchmark
loss experience, as described in the next section, captures the
strenuous economic conditions required by the 1992 Act without adding
more economic variables. Minimizing the number of variables used to
define economic conditions is responsive to the comments of both Fannie
Mae and Freddie Mac, who argued against unnecessary complexity.
(e) Purchase vs. Refinance Loans
MRAC suggested that OFHEO take loan purpose into account. OFHEO
considered whether this distinction should be included as a variable,
but has proposed a stress test that does not distinguish between loans
made for the purpose of purchasing and loans made for the purpose of
refinancing property. OFHEO has found insufficient basis to distinguish
between the risks of loans for purchases and loans for refinancing.
Furthermore, OFHEO prefers not to create capital incentives based on
loan purpose, except as required by statute (e.g., the occupancy status
distinction).
(f) Credit Scores
OFHEO does not propose to follow the recommendation of MRAC to use
mortgage borrower credit quality considerations as explanatory
variables. OFHEO is aware that the mortgage industry is moving toward
risk-based loan pricing based, in part, on mortgage credit scores that
rely heavily on borrower credit ratings.\109\ OFHEO is studying the use
of credit scores by the Enterprises, and the potential for impact on
stress test credit losses, but does not believe that it is appropriate
to consider these in the stress test or to use them to estimate the
models. First, it would be difficult, if not impossible, to reasonably
relate credit risk differences based upon credit scores to the
benchmark loss experience, because credit-scoring data are not
available for benchmark era loans.\110\ Second, the proposed stress
test is designed to reasonably relate starting the performance of
mortgage portfolios to the benchmark loss experience based upon loan
characteristic differences referenced in the 1992 Act, which do not
include measures of borrower creditworthiness.\111\
---------------------------------------------------------------------------
\109\ The most widely used measure of borrower creditworthiness
is a composite score developed by Fair Isaac Corporation, commonly
referred to as a ``FICO score.''
\110\ Archives at the credit repositories only go back to the
late 1980s, and, even there, records are not complete.
\111\ The fact that OFHEO does not consider differences of
credit risk by credit scores in the proposed stress test does not
limit the ability of the Enterprises to to make use of credit
scores. The Enterprises may further stratify the risk
classifications used by OFHEO in the proposed stress test, for
purposes of internal capital allocation and guarantee pricing. For
example, after determining the required regulatory capital for a
particular product class the Enterprises may, if they choose,
allocate the required capital among purchases of that product
according to borrower credit scores, for internal purposes. Thus,
the dimensions on which the Enterprises choose to develop risk-based
guarantee pricing are not limited by stress test risk
classifications.
---------------------------------------------------------------------------
[[Page 18136]]
(vii) Additional Multifamily Explanatory Variables
Understanding the choice of explanatory variables for the
multifamily default/prepayment model requires understanding the way in
which default and prepayment equations are organized. The stress test
uses two default equations, to distinguish between different
multifamily lending programs, and five prepayment equations, to
distinguish between different product types. The multifamily model
allows these various default and prepayment equations to interact with
each other to provide appropriate default and prepayment rate
projections for all multifamily loans, throughout the stress period.
One of the two default equations is for purchases of newly
originated loans (cash purchases),\112\ and the other is for negotiated
swaps of seasoned loan pools for mortgaged-backed securities
(negotiated purchases). This separation allows the stress test to
account for differences in loan quality across the two programs. The
Enterprises may take lower quality loans and properties in their
negotiated purchase programs than in the cash purchase programs, but
require significant credit enhancements from the seller/servicers to
compensate.
---------------------------------------------------------------------------
\112\ Cash-purchase programs may involve delivery of loans for
cash or for mortgaged backed securities. They are called ``cash''
programs because they involve the purchase of individual loans under
published underwriting guidelines and pricing.
---------------------------------------------------------------------------
The five prepayment equations used to accommodate product-type and
product life-cycle differences allow the proposed stress test to
account for the effects of loan characteristics, such as yield-
maintenance provisions,\113\ adjustable interest rates, and balloon
terms. It is more important to capture the unique features of balloon
mortgages in the multifamily business than it is in the single family
business because balloons make up the majority of multifamily
portfolios. The five prepayment equations are for: (1) All fixed-rate
loans in the yield-maintenance period; (2) fully-amortizing fixed-rate
loans after yield maintenance requirements; (3) fixed-rate balloon
loans after the expiration of yield-maintenance requirements (but prior
to maturity); (4) all ARM loans (prior to maturity for balloon ARMs);
and (5) all balloon loans (with fixed or adjustable interest rates) at
and after the maturity year.
---------------------------------------------------------------------------
\113\ A yield maintenance provision permits prepayment, but
requires the borrower to pay penalties to compensate the lender or
investor for lost interest until the yield maintenance period
expires.
---------------------------------------------------------------------------
To see how these prepayment equations work together, note, for
example, that fixed-rate balloon loans have three relevant time
periods: first is ``in-yield maintenance,'' the time when the yield
maintenance terms apply; second is ``post yield maintenance,'' the
period after the yield maintenance term expires and prior to loan
maturity; and third is ``post-balloon,'' the period starting when the
loan is due in full.\114\ For loans that extend to and beyond the
balloon point,\115\ OFHEO proposes a separate prepayment equation,
which is referred to as a ``payoff'' equation because it is no longer
possible to ``prepay'' loans on or after the balloon date.
---------------------------------------------------------------------------
\114\ Balloon loans with adjustable interest rates (rather than
fixed coupon rates) do not have yield maintenance terms, so they
only have two relevant periods--pre- and post-balloon.
\115\ After the balloon maturity date, the Enterprises may
permit loan extension.
---------------------------------------------------------------------------
(a) Explanatory Variables in the Two Multifamily Default Equations
The two multifamily default equations are similar except in two
respects. First, the equation for cash purchases makes adjustments for
loans purchased in original multifamily programs to distinguish them
from more recent programs. Second, the negotiated purchase loan
equation has an adjustment factor for loan programs that obligate the
seller to repurchase loans when they are delinquent for 90 days. These
distinctions will be discussed in the context of each explanatory
variable.
(1) Joint Probability of Negative Equity and Negative Cash Flow
As with single family loans, one of the most important factors
affecting multifamily loan default is borrower equity. When the value
of the property is less than the value of the mortgage, the borrower,
by defaulting, can effectively ``sell'' or ``put'' a mortgage back to a
lender at the value of the underlying property. However, as recognized
by the ANPR commenters, there is a second consideration for commercial
properties (including multifamily properties)--cash flow from the
property. Even though equity is zero or negative, the borrower does not
have an economic incentive to default as long as cash flows are
positive.
The stress test includes a default option valuation variable that
allows for consideration of the cash flow position of the property,
while also considering the borrower's equity position. A value for this
variable, referred to as the joint probability of negative equity and
negative cash flow, is calculated for each loan in each observation
period. It measures the potential value of ``putting'' the mortgage to
the lender and investor through default, given that both equity and
cash flow are important.\116\
---------------------------------------------------------------------------
\116\ The equity and cash flow positions of a property are
positively correlated. The joint probability of negative equity and
negative cash flow variable used in the proposed stress test
captures this relationship.
---------------------------------------------------------------------------
As shown in section D. 4. a. i., Joint Probability of Negative
Equity and Negative Cash Flow, of the Technical Supplement, the joint
probability of negative equity and negative cash flow for a project is
the probability of having both LTV greater than 1.00 and DCR less than
1.00. The proposed stress test uses loan amortization schedules, rental
inflation, vacancy rates, and interest rates to update LTV and DCR,
which are then used to update the joint probability variable values.
(2) Original Versus Current Loan-Purchase Programs
OFHEO faced the issue of what, if any, adjustment should be made in
the model to distinguish between loans purchased under original cash-
purchase programs (purchased pre-1988 for Fannie Mae and pre-1992 for
Freddie Mac) and current programs. As noted by Freddie Mac, the
Enterprises computed both DCR and LTV differently for loans purchased
under original programs than they compute those ratios today for
current purchase programs. OFHEO recognizes that in the 1980s it was a
common appraisal practice to adjust actual rents (and therefore net
operating income) upward by an estimate of annual inflation and to use
optimistic vacancy rate assumptions. This practice resulted in an
overstatement of actual DCR and LTV values at the time of loan
origination. Current practice does not allow for such inflation
adjustments of projected rents, and factors minimum levels of
anticipated vacancies into property valuation, even if the property is
fully rented at the time of loan origination.
In addition to the overstatement of net income, original
multifamily cash-purchase programs at the Enterprises had other
significant weaknesses perhaps because the Enterprises only began
purchasing conventional multifamily loans in 1983 and did not have
experience with the differences from single family lending. Even
controlling for the overstatement of rents and for changes in tax laws
in 1986 that depressed real estate values, these weaknesses led to
extraordinarily high loss rates. OFHEO views these large losses, to a
large extent, as nonrecurring startup costs attributable
[[Page 18137]]
to inefficiencies involved in learning a new business. For these
reasons, OFHEO believes that the Enterprises' multifamily lending
programs in the early and mid-1980s are so different from the current
programs that it would be inappropriate to consider those early loans
to be the same type of mortgage product as the multifamily loans that
are made today.
The stress test accounts for the difference in the older loan
programs and the newer programs in two ways. First, the stress test
adjusts the origination DCRs and LTVs of original cash purchase loans
to remove the estimated annual inflation factors and restate those
ratios as they would be calculated by the Enterprises in their current
program purchases.\117\ Second, the stress test includes a variable in
the default equation that distinguishes between original and current
cash purchase programs. This variable results in higher levels of
default on original cash purchase loans than on newer loans.
---------------------------------------------------------------------------
\117\ OFHEO found that loans acquired in negotiated swap
arrangements in the early and mid 1980s were highly seasoned and had
low default rates. They therefore did not appear to include the
inflation factor evident in cash purchases. Therefore, OFHEO does
not adjust DCRs and LTVs for loans in negotiated purchase pools.
---------------------------------------------------------------------------
A significant consideration in OFHEO's proposal to distinguish the
original cash purchase loans from loans purchased under current
programs was that failing to make that distinction would create a
relatively more severe (and far less) loss experience for multifamily
loans than the benchmark loss experience creates for single family
loans.\118\ In OFHEO's view, imposition of such extreme levels of
default upon the Enterprises' multifamily loans would be contrary to
the intent of the 1992 Act that rates of default and severity be
``reasonably related'' to the benchmark loss experience. It is also
possible that basing stress test losses on average default rates of
original cash-purchase loans would result in an implied marginal
capital requirement so high as to create an inappropriate disincentive
to engage in new multifamily lending.
---------------------------------------------------------------------------
\118\ The relationship of multifamily default rates to the
benchmark experience is discussed later in section III. A. 7. c.,
Relating Multifamily Mortgage Performance to the Benchmark.
---------------------------------------------------------------------------
(3) Depreciation Write-offs and Tax Law Changes
In the absence of a price index for multifamily properties, the
stress test captures most of the changes in property value by updating
DCR and LTV according to changes in rents, vacancies, and interest
rates. However, changes in DCR and LTV that are due to other factors
are not captured in these procedures. The most important missing factor
is the tax benefit afforded to owners of investment real estate through
depreciation write-offs. ACB commented that depreciation allowances
have important effects on property cash flows. OFHEO recognizes this
fact and that the allowances also have important effects on capital
gains at the time of property sale. The tax value of depreciation
write-offs significantly influences the return from multifamily
property investments and, consequently, the default risk of multifamily
mortgages.
OFHEO agrees with Freddie Mac that tax law changes affecting
multifamily default rates during the 1980's should be taken into
account, but that OFHEO should not speculate on the effect of potential
legislative or other governmental actions during the stress period. The
proposed stress test incorporates an index that measures the value of
depreciation write-offs for a new investor. It measures changes in
quality due to changes in write-offs and allows OFHEO to reflect the
effects of such changes on mortgage defaults historically. The actual
index value used in the stress test is an approximation of expected
values throughout the stress period.\119\ It is calculated based on
depreciation rules and tax rates as they existed in 1997, with no
adjustments for movements in interest rates since that time, or for the
interest-rate shocks that will occur in the stress test. The tax rules
governing depreciation allowances have the largest impact on the value
of this variable. These rules changed significantly in 1986, but have
not changed significantly since. Because the historical database
included many loans originated before the tax rule change, OFHEO
allowed the value of this explanatory variable to vary for purposes of
estimating the statistical equations for multifamily mortgage default.
However, due to the subsequent stability in those rules, OFHEO proposes
to hold the value of this variable constant throughout the stress test.
If the applicable tax rules change in the future, or if OFHEO believes
that there are other reasons for either changing the specified value
for the stress test or allowing its value to change throughout the
stress test, OFHEO will initiate a new rule making process. However, as
recommended by Freddie Mac in its ANPR comments, OFHEO will not
speculate about tax law changes that might occur during the stress
period. Due to data restrictions, the depreciation-allowance is only
included in the cash-purchase default equation.\120\
---------------------------------------------------------------------------
\119\ The stress test does not capture actual depreciation
allowances for borrowers. Enterprise databases do not include the
year of property purchase. Therefore, the exact depreciation rules
affecting cash flows and investment value to existing owners are
unknown. Even on newly constructed projects, the Enterprises
generally do not purchase the mortgage until target occupancy rates
are met, which may be some time after origination. For these
reasons, it would be extremely difficult to determine the actual
value of depreciation write-offs to current owners. Although the
value to current owners affects the owner's cash flow, the value to
potential purchasers (which would be based upon current appreciation
rules) affects property value and the owner's equity in the
property. Therefore, this explanatory variable for depreciation
write-offs helps to reflect more accurately the true LTV of the
mortgage.
\120\ See section D. 4. a. ii., Construction of the JPt Variable
of the Technical Supplement for details.
---------------------------------------------------------------------------
(4) Loan Programs with Seller/Servicer Repurchase Features
Some Enterprise multifamily loan programs require seller/servicer
repurchases of loans that become 90-days delinquent. For these programs
a 90-day delinquency event is effectively a default, while for all
other loans, default means a property loss event (short sale, note
sale, third-party sale or foreclosure). To account for this difference
when estimating the statistical model, OFHEO applied, as an explanatory
variable, the ratio of 90-day delinquencies to full defaults. This
treatment is important because the rate of 90-day delinquency events is
always higher than the default rate for property loss events, and the
loss severity for 90-day delinquencies is lower. By including this
ratio, and thus including loans with the 90-day delinquency
terminations, OFHEO was able to estimate a negotiated-purchase default
equation based on a much larger data set than would have been possible
otherwise.
(5) Balloon and ARM Payment Shock Risk
Following HUD's suggestion, OFHEO analyzed defaults of Enterprise
balloon loans at the balloon point. As a result, OFHEO proposes to give
additional weight to the joint probability of negative equity and
negative cash flow variable for balloon loans that survive to the year
of balloon maturity. This extra weighting takes into account the
increased risk that mortgages with weak financials will default as the
balloon point approaches. Also, interest rate movements may create
payment shock (change in the periodic mortgage payment) in the post-
balloon period, which affects the probability of default. The stress
test accounts for the effect of
[[Page 18138]]
this shock directly through adjustments to effective DCR in the post-
balloon period. These adjustments then affect the joint probability of
negative equity and negative cash flow, reflecting the fact that the
decision to default or payoff is no longer a function of the original
mortgage coupon rate, but of the prevailing market rates at the time of
balloon expiration. In sum, the stress test reflects that the value of
the default (``put'') option, as measured through the joint probability
variable, becomes more significant for default rates in the post-
balloon period because there is increased pressure on the borrower to
either default or refinance the property.
ARMs also experience payment shock because of changes in market
interest rates. ARM payment shock occurs periodically during the term
of the loan, and ARMs continue to amortize after the payment shock,
according to the original contract term. The ARM prepayment equation in
the stress test accounts for these periodic changes in interest rates.
In contrast, the payment shock for a fixed-rate balloon loan does not
occur until the balloon point. Some loans in Enterprise portfolios are
ARMs with a balloon maturity. These loans have payment shock every year
and also at maturity. The proposed stress test models the annual
changes in their DCRs resulting from changes in mortgage coupon rates
and then adds an additional balloon shock through the additional weight
given to the joint probability variable in the post-balloon period.
(6) Loan Size
The stress test does not include a variable for loan size. S&P
explained that it bifurcates commercial loan pools into two parts to
calculate credit loss potential--the largest loan, and all other loans
in the pool. S&P assumes 100 percent risk of default on the largest
loan and average risk of default on the other loans. This approach is
designed to recognize the uneven dollar credit loss risk inherent in
pools that contain loans that are large relative to the total size of
the pool. Credit risk for the pool is then estimated by S&P to be the
sum of estimated credit risk on each part. S&P did not specifically
recommend that OFHEO adopt this approach in the stress test.
OFHEO agrees that S&P's methodology is appropriate for analyzing
differential impact of large and small loans on potential credit losses
in mortgage security pools. However, no one multifamily loan default
could have a significant impact on total losses or capital for either
Enterprise. For that reason, OFHEO decided not to propose any measure
of loan size as an explanatory variable in the multifamily default/
prepayment model.
(b) Explanatory Variables in the Five Multifamily Prepayment Equations
As explained above, the multifamily model uses five loan prepayment
equations to identify unique product type and life-cycle
characteristics. This approach is consistent with Freddie Mac's and
MRAC's comments on accounting for mortgage product types and terms in
the default and prepayment models. There are some differences in
explanatory variables across these five equations, which are discussed
below.
(1) Prepayment Option Value
As discussed earlier, OFHEO proposes to use the relative interest
rate spread to measure the prepayment option value (mortgage premium
value) for prepayments. The relative spread is the ratio of the
difference between the coupon rate and the current market interest rate
to the coupon rate. To account for the asymmetry of effects from
increases and decreases in interest rates, the spread is split into two
variables.\121\ One is active if current market interest rates are
above the mortgage coupon rate, and the other is active if current
market rates are below the mortgage coupon rate. Decreased interest
rates increase refinancing speeds. Increased interest rates decrease
both normal refinancings and cash-out refinancings. Cash-out
refinancings are refinancings in excess of the outstanding
indebtedness. They are used to achieve a desired debt-to-equity ratio
in the property as explained below in the discussion of current LTV.
Relative spread variables appear in all prepayment equations except for
the balloon and post-balloon payoff equations. At balloon maturity, all
spreads become irrelevant, because borrowers are contractually
obligated to pay off or refinance the property.
---------------------------------------------------------------------------
\121\ Such explicit bifurcation is not required for the single
family prepayment equations because the categorical nature of the
spread variable used there allows for asymmetric effects.
---------------------------------------------------------------------------
For the ARM prepayment equation, the relative spread variable is
calculated by comparing the coupon rate to the current market rate on
fixed-rate loans, rather than to the market rate for ARMs. This
approach accounts for any incentive to refinance into a fixed-rate
loan. Because there are no yield-maintenance terms or special
incentives to refinance ARM loans when interest rates fall, the stress
test includes one spread variable that captures both increases and
decreases in interest rates. In addition, the stress test does not
distinguish between life-cycle periods for ARMs; just one prepayment
equation is estimated.
(2) Current LTV
Another important issue in modeling multifamily loans is the
propensity of investors in multifamily properties to refinance
mortgages over time to increase their debt (leverage) ratios, and thus
increase returns on invested equity.\122\ To capture the borrowers'
ability to qualify for a new loan and the incentive to adjust debt-to-
equity ratio, the proposed stress test includes current LTV as an
additional explanatory variable. If the current LTV falls, investors
have more incentive to prepay and are more likely to find a lender
willing to refinance the property.
---------------------------------------------------------------------------
\122\ See Jesse M. Abraham and H. Scott Theobald, ``Commercial
Mortgage Prepayments,'' in Frank Fabozzi and David Jacob, The
Handbook of Commercial Mortgage-Backed Securities, New Hope, PA:
Frank J. Fabozzi Associates, 55-74 (1997).
---------------------------------------------------------------------------
(3) Prepayment Option Value in the Yield-Maintenance Period
During the yield-maintenance period, borrowers may prepay, but they
must continue to provide the contractual yield until the yield-
maintenance period expires. Thus, a prepayment in the yield-maintenance
period can be expensive, particularly in the early years of a mortgage.
The more years to go in the yield-maintenance period, the greater the
fee.\123\ To capture the declining financial cost of prepayment
throughout the yield-maintenance period, OFHEO proposes a variable
measuring years remaining until the end of the yield-maintenance
period. This variable appears in the prepayment equation for fixed-rate
loans in the yield-maintenance period.\124\
---------------------------------------------------------------------------
\123\ Because this effect runs counter to the effect of the call
option value, OFHEO researched the possibility of a joint effect of
the years-to-go and the rate drop variables. The fixed effects of
the years-to-go variable proved to be a better predictor of actual,
historical prepayments during yield maintenance periods.
\124\ For loans with true prepayment prohibitions, or ``lock-
outs,'' the variable is set equal to the maximum number of lockout
years throughout the lockout period. See section 3.5.4.3,
Procedures, of the proposed Appendix to 12 CFR part 1750, subpart B
for details.
---------------------------------------------------------------------------
(4) Prepayment Option Value in the Pre-Balloon Period
During the pre-balloon period, borrowers are uncertain about the
level of market interest rates at the future balloon point. Hence,
borrowers may be willing to pay in order to lock into a favorable
interest rate, rather than take
[[Page 18139]]
their chances with possible adverse interest rate movements. This risk
aversity with respect to interest rate movements prior to the time of
balloon maturity gives rise to an additional financial value from early
prepayment. OFHEO proposes two explanatory variables to capture the
effect of risk aversity on prepayment rates in the pre-balloon period.
They measure the additional effects of the primary prepayment option
variable-relative spread-when it is in the money (market interest rates
are lower than the mortgage coupon rate).
The first variable provides an additional effect for interest rate
drops in the year immediately prior to the balloon year, and the second
provides for a separate, additional effect for interest rate drops in
the second year prior to the balloon year. These two variables allow
for increased incentives to refinance if the prepayment option is in
the money in the period leading up to balloon expiration. They capture
the risk aversity of borrowers with respect to future interest rate
changes as balloon maturity approaches.
(5) Balloon and Post-Balloon Payoffs
HUD commented that OFHEO should model the value of the refinancing
option at the balloon point on balloon mortgages because the lender
often has a contractual obligation to refinance at the borrower's
option. OFHEO agrees that payoffs at the balloon point are different
from prepayments before the maturity date, but has found that the
lender generally does not have an unconditional contractual obligation
to provide new funding if the borrower requests it. Payoff of the
balloon loan (generally by new borrowing to refinance the property) is
contractually required at term. If the borrower is successful at
finding new financing at that point, the event that appears in
Enterprise records is a payoff of the original loan and not a
prepayment. Despite the contractual requirement of balloon payoff, not
all loans terminate at the balloon point.\125\ Generally, balloon loans
are extended beyond the maturity date because, although the property
has weak financials, lenders are unwilling to initiate foreclosure on
loans that have been making payments at the original coupon rate. To
capture the ability of multifamily borrowers to obtain new loans at
balloon expiration, and, therefore, to pay off the original mortgage,
the model includes a variable similar to the joint probability variable
used in the default equations--the joint probability that current DCR
and LTV values are sufficient to qualify for a new mortgage. This is
the only variable used in the pay-off equation for balloon mortgages,
and it is based on minimum qualification criteria for multifamily
mortgages, LTV 0.80 and DCR 1.20.
---------------------------------------------------------------------------
\125\ See Elmer and Haidorfer, ``Prepayments of Multifamily
Mortgage-Backed Securities,'' The Journal of Fixed Income, March
1997, 50-63 (pointing out that not all loans terminate at balloon
point); Abraham and Theobald, op. cit. (referring to this phenomenon
as extension risk). OFHEO confirms the existence of post-balloon
loans in Enterprise portfolios.
---------------------------------------------------------------------------
(6) Effect of Fixed-Rate Loan Interest Rates on ARM Prepayments
A final variable included in the ARM prepayment equation is the
market rate on fixed-rate loans. This variable accounts for incentives
to refinance ARM loans into fixed-rate loans to avoid future
uncertainty regarding interest rate movements. If the FRM rate is high,
borrowers expect interest rates to drop in the future and are likely to
delay prepayment of ARMs. Likewise, when interest rates are low--
regardless of the spread between FRM and ARM rates--there is an
incentive to refinance into a fixed-rate product to avoid potential
increases in future interest rates.
6. Loss Severity
Loss severity is the net cost to an Enterprise of a loan default.
The three major cost categories are loss of loan principal transaction
costs at both foreclosure and disposition, and asset funding costs
throughout the process. The net cost is determined by crediting against
these costs the revenues associated with the defaulted loan. The major
revenues are proceeds from the property sale and from mortgage
insurance or other forms of credit enhancement.
In determining how to model loss severity in the stress test, OFHEO
considered the following issues:
1. what general approach to take in modeling loss severity,
2. whether the stress test should model individual cost and revenue
elements of loss severity or model severity as one single measure,
3. what explanatory variables should be included explicitly in
modeling loss severity, and
4. an appropriate house price index for real estate owned (REO)
properties.\126\
---------------------------------------------------------------------------
\126\ REO properties are properties acquired as a result of
foreclosure or similar action.
---------------------------------------------------------------------------
a. General Approach to Modeling Loss Severity
In the ANPR, OFHEO discussed four general approaches to estimating
the separate effects of explanatory variables on loss severity. One
approach is to use a multivariate statistical model to estimate the
separate effects of explanatory variables on total loss severity rates.
A second approach is to use statistical models relating the individual
elements of loss severity to explanatory variables. A third approach
would set fixed parameters for the elements of loss severity
(foreclosure costs, carrying costs, and sales prices), while allowing
final loss severity rates to vary based on other factors such as the
presence of private mortgage insurance. A fourth, relatively simple
approach would be to assume that all defaulted loans face a fixed and
equal level of loss severity.
(i) ANPR Comments
ACB and MRAC encouraged OFHEO to use a multivariate statistical
model of loss severity. ACB, apparently assuming the stress test would
include a statistical model of defaults, stated that ``[i]t is not a
rational allocation of resources to develop a sophisticated model of
mortgage defaults and then to apply a rule-of-thumb percentage to the
unpaid principal balances.'' S&P described its use of data from the
Great Depression as the basis for stress tests it uses to rate single-
family mortgage pools. Freddie Mac recommended that OFHEO use average
loss severity rates from the benchmark loss experience, adjust them to
account for the stress test interest rate environment, and apply
additional adjustments for various property types.
(ii) OFHEO's Response
OFHEO believes that a statistical model is the best approach to
take into account loan seasoning and the dynamic nature of economic
changes in the stress period. OFHEO agrees with ACB that it would be
inappropriate to develop a sophisticated default model and then to
apply a rule-of-thumb percentage to the UPB to determine loss severity.
At the same time, OFHEO recognizes that developing statistical models
of each loss element is unnecessarily complex. Based on its analysis of
the available information, OFHEO proposes a two-part model for single
family loss severity: a statistical equation for loss of loan principal
and fixed parameters for the other cost elements. Specifically, the
statistical model developed by OFHEO estimates loss of loan principal
as a function of loan seasoning-updating the original LTV using HPI
growth rates and loan amortization. For multifamily loss severity,
OFHEO proposes to use only fixed cost element values. The rationale for
this is explained below under section III. A.7., Relating Losses to the
Benchmark Loss Experience.
[[Page 18140]]
The approach outlined by S&P would not be appropriate for OFHEO's
stress test because it does not adjust for loan seasoning or provide
for a reasonable relationship to the benchmark as required by the 1992
Act. However, consistent with the S&P approach, the stress test does
provide for a greater than average drop in house prices for foreclosed
properties. As discussed below, under section III. A.6. b., Elements of
Loss Severity Modeled, the stress test uses a statistical equation to
model the expected decline in values on foreclosed properties, which
will be greater than the decline in property value associated with HPI
assumptions used in the stress test. In addition, as discussed later
under section III. A.7., Relating Losses to the Benchmark Loss
Experience, the stress test adds an extra loss factor to relate stress
test property value loss to the actual experience of the four-State
benchmark.
OFHEO agrees that Freddie Mac's recommended approach is simpler
than using a statistical model. However, an empirically based
statistical model is more versatile and flexible, allowing the stress
test to reflect loss severity rates appropriate for each Enterprise's
mix of loans and the stress test interest rate environment. OFHEO
proposes a hybrid approach that retains the simplicity of fixed cost
factors for most severity elements, while developing a more sensitive
measure of property value, the element most affected by pre-stress test
loan seasoning.
OFHEO does not propose at this time to take property type
differences into account in stress test loss severity rates, as
suggested by Freddie Mac. Although OFHEO finds higher loss severity
rates for investor-owned properties, accounting for this effect would
increase significantly the number of loan group records used for
starting books of business in the stress test. Given the small
percentage of Enterprise portfolios that investor-owned loans comprise,
OFHEO felt that the added complexity was not justified by the benefits
of calculating severity rates for owner-occupied and investor-owned
single family loans separately. Therefore, OFHEO does not propose to
apply risk multiples for investor-owned properties in determining loss
severities. Rather, the single set of cost elements used in the stress
test are determined by Enterprise experience with all single family
property types combined.
b. Elements of Loss Severity Modeled
In addition to asking whether OFHEO should use a statistical model
of loss severity, the ANPR asked whether the stress test should model
loss severity as a single value or model the various cost and revenue
elements of severity separately.
All ANPR commenters favored, at varying levels, an element-by-
element analysis. The VA recommended that the stress test model the
amount and timing of both the cost and the revenue elements of loss
severity to provide more accurate estimates of Enterprise cash flows.
HUD recommended that the loss severity model include certain individual
cost elements, all of which would be valued separately by the proposed
severity module. NAR stated that ``the modeling of loan loss severity
should only include those factors that are independent of incidence of
default'' and emphasized the importance of modeling time in default
separately. In contrast, Freddie Mac stated that defaults and severity
are products of the same underlying characteristics and economic
factors. Freddie Mac suggested that stress test severity calculations
differentiate loans by original LTV and coupon class and by product
type distinctions. In addition, Freddie Mac favored using the rate of
loss of principal balance from the benchmark loss experience.
ACB supported using a sophisticated model of loss severity, which
would, presumably, require breaking down severity into its constituent
parts for analysis and modeling. MRAC suggested separate analysis of
the elements of loss severity, including the estimated sale proceeds,
holding time, monthly holding costs, and costs of sale.
OFHEO agrees with the commenters that the stress test should model
individual cost and revenue elements separately, rather than model them
together as a single cost category. Such an approach allows the stress
test to model the interrelationship of those elements that
significantly effect loss severity. Accordingly, OFHEO proposes to
model elements in three principal groupings: (1) loss of loan principal
balance, (2) transaction costs (e.g., expenses related to foreclosure,
and property holding and disposition expenses), and (3) funding costs
on non-earning assets. OFHEO believes that measuring elements in these
groupings is necessary to accommodate differences in the timing of
various elements of loss severity and differences in the pre-stress
test seasoning of loans. Each cost or revenue factor is applied at one
of the following three points in time (each in terms of months from
date-of-default): time of loan repurchase (for loans in security pools)
or bad-debt write off (for retained loans); time of foreclosure
completion; and time of foreclosed property disposition.
In addition, consistent with Freddie Mac's comment, OFHEO's
proposed loss severity calculations differentiate by LTV and coupon
class. They also include product distinctions where those distinctions
involve FHA/VA insurance, interest rates and amortization terms. The
amount of the loss of loan principal balance is sensitive to loan
amortization. Because 15-year mortgages amortize relatively early and
more quickly, their predicted losses are much less than those on
otherwise comparable 30-year mortgages.
(i) Loss of Principal Balance
A critical element of loss severity is loss of loan principal
balance, i.e., the difference between the outstanding principal balance
on the loan at the time of default and the sale price of the foreclosed
property. This loss occurs because of general declines in local housing
values, the depreciation of the individual property, and/or discounts
required to sell properties with ``foreclosure'' labels. To calculate
this loss, the stress test uses a statistical model of the historical
relationship between actual loss of principal balance on loans that
have defaulted and the loss of principal balance predicted solely by
calculating amortization on the loan and updating the property values
with the HPI. Sale proceeds are then calculated as UPB minus the
estimated loss of principal balance. Proceeds vary with differences in
house-price appreciation and loan terms.
(ii) Transaction Costs
The stress test includes two transaction cost elements in loss
severity calculations: foreclosure/legal expenses, and property holding
and disposition costs.\127\ Property holding and disposition costs are
combined in the proposed stress test because they are both expensed at
the time of property disposition. OFHEO proposes to use averages of
these cost elements--in percent of outstanding principal balance--from
all Enterprise experience with foreclosure and REO properties.
---------------------------------------------------------------------------
\127\ Legal expenses are dominated by foreclosure costs, but
they also include costs associated with gaining releases from
borrower bankruptcy stays and property evictions.
---------------------------------------------------------------------------
OFHEO did not follow Freddie Mac's recommendation to use all cost
elements directly from the benchmark loss experience for transaction
costs, because the stress test is national in scope. Therefore, it is
appropriate to have a national blend of institutional factors such as
foreclosure costs, property management fees, and sales
[[Page 18141]]
expenses, rather than the four-State blend from the benchmark
experience.
(iii) Funding Costs
Funding costs are considered an element of loss severity because
the Enterprises must fund non-earning assets: first the defaulted
loans, and then the REO properties. In its ANPR comments, Freddie Mac
suggested that funding costs should be measured at the mortgage
interest rate for the period from date of default to foreclosure
completion. OFHEO agrees that the stress test should model funding
costs. However, Freddie Mac's recommended approach ignores funding
costs during the REO time period and would provide inaccurate measures
of funding costs during the delinquency/default period. In the down-
rate scenario of the stress test, using the mortgage coupon rate for
funding costs would overstate funding costs, while in the up-rate
scenario it would understate funding costs.
With one exception, the stress test measures asset funding costs
through present-value discounting techniques, rather than computing
explicit interest charges. Therefore, all severity elements are
discounted by a cost-of-funds rate to produce the present value of each
element in the month of default, regardless of when it may occur after
that date. Cash flow discounting provides a consistent method of
accounting for all timing issues involving cash flows from mortgage
default to property disposition.
The one exception to the rule of calculating funding costs through
present-value discounting techniques is the explicit cost of covering
interest passed through to investors in securitized loans (mortgage-
backed securities). These passthroughs occur for the first four months
of loan delinquency, during which time the stress test uses the
passthrough rate (the interest rate paid to holders of the securities)
to calculate the asset funding cost. After the fourth month, when the
loans have been repurchased from security pools and placed in
Enterprise retained portfolios, the stress test treats these defaults
identically to defaults in retained portfolios.
(iv) Factors Not Modeled
ANPR commenters suggested several explanatory factors that are not
included in the proposed single family loss severity model. These
include distinctions based on State foreclosure laws, household
liquidity, and the presence of private mortgage insurance.\128\
---------------------------------------------------------------------------
\128\ Although private mortgage insurance is not an explanatory
variable, proceeds from such insurance are accounted for in the
severity calculation.
---------------------------------------------------------------------------
(a) State Foreclosure Law Differences
Freddie Mac suggested that OFHEO not make State-level distinctions
in loss severity calculations, explaining that attributing
``differences in loss rates by states would approach undue intrusion
and inappropriate micromanagement of the Enterprises.'' In contrast,
NAR recommended that OFHEO make State distinctions.
Although foreclosure time-frames and costs may vary based on State
law and practice, OFHEO agrees with Freddie Mac that it would be
inappropriate to model State-level differences. First, these
differences do not represent loan characteristics, and, therefore,
under OFHEO's approach to selecting variables to apply in the stress
test, they are not appropriate. Second, if OFHEO were to allow for
State-level differences in credit costs, the stress test would,
essentially, be establishing State-specific capital requirements based
upon nuances of State law. OFHEO would need to monitor developments in
the many different State laws over time to adjust the parameters of the
stress test. Third, the fact that the stress test uses loan data
aggregated at the Census division level means that much of the
variability in foreclosure costs observed at the State level
disappears.
(b) Independence of Loss Severity Rates From Default Rates
Freddie Mac commented that default and loss severity are products
of the same underlying factors, most particularly original LTV and
property value appreciation over the life of the mortgage. NAR
recommended that the loss severity model ``only include those factors
that are independent of the incidence of default.'' OFHEO agrees with
Freddie Mac on this point, because OFHEO's research indicates that loan
seasoning has an important impact upon severity rates that is
independent of its impact on defaults. The use of loan seasoning in the
stress test reflects differences in loss severity across loans. This
approach is also consistent with NAR's comment, because estimating the
impact of seasoning on loss severity independently from its impact on
defaults avoids duplicating seasoning's effect on credit losses.
(c) Household Liquidity
NAR stated that liquidity of the household under stress is an
important factor in the loss severity equation. OFHEO notes that for
the single family loss severity analysis, the stress test considers
housing-related liquidity of a household through loan seasoning. That
is, updating the LTV provides some indication of the ability of
borrowers to sell or borrow against their properties in order to
provide liquidity. However, the stress test does not account directly
for non-housing wealth or liquidity of borrowers. It is unclear how
these factors could be measured or estimated accurately.
(d) Private Mortgage Insurance
NAR also commented that the presence of private mortgage insurance
is a variable that can influence the time to foreclosure and therefore,
presumably, holding costs. OFHEO, however, has found insufficient
evidence that the presence of mortgage insurance has any meaningful
impact on foreclosure time. Both Enterprises submit their own
foreclosure time guidelines to seller/servicers, which are independent
of the presence of mortgage insurance. Accordingly, the presence of
private mortgage insurance is not included as a variable in the loss
severity equations.
This issue is distinct from the question of how OFHEO should
account for private mortgage insurance proceeds in the loss severity
calculations. Several commenters noted that the loss severity
calculation should deduct mortgage insurance proceeds from losses on
loans covered by such insurance. OFHEO agrees that the loss severity
calculation should account for mortgage insurance proceeds. This issue
is discussed extensively in section III.C., Mortgage Credit
Enhancements.
c. REO House Price Index
In the ANPR, OFHEO asked what price index would be appropriate for
REO properties. The question arose because defaulted loans generally
have lower house-price appreciation rates than the market average,
which is captured by HPI growth over time. After considering the ANPR
comments and OFHEO's own research, OFHEO proposes an equation to relate
actual declines in value for REO properties to changes in the HPI. This
approach, which is described in section 3.5.3.3.3.1, Calculate Proceeds
from Property Sale, of the Regulation Appendix, provides the
information needed to predict accurately the loss of loan principal
balance in loss severity calculations, but avoids the added complexity
of creating a separate index.
All five commenters that addressed this issue recognized that,
without
[[Page 18142]]
adjustment, the HPI would not provide an adequate measure of REO price
changes. However, none recommended creation of a separate REO index.
Four commenters (MRAC, ACB, VA, and Freddie Mac) recommended modifying
the general price index. MRAC suggested that a general HPI be used in
conjunction with analysis of variances of prices to determine whether
foreclosure prices have experienced slower appreciation or greater
depreciation than the market average. ACB suggested that, rather than
developing an REO price index, OFHEO study the ``left tail'' of the
distribution of house prices in general. The term ``left tail'' refers
to those houses with the smallest appreciation rates. S&P provided to
OFHEO the rates of property value loss for foreclosures during the
Great Depression.
The proposed approach incorporates a statistical model based upon
an analysis like that suggested by MRAC and ACB. The model predicts how
far into the left tail each REO property value can be expected to be,
relative to the outstanding mortgage balance, throughout the stress
period. OFHEO's proposed approach essentially follows the specific
recommendations of MRAC and ACB for modification of the HPI.
The VA suggested using a general house price index, re-weighted to
capture the regional distribution of REO properties. OFHEO agrees that
regional differences in REO appreciation rates should be captured. The
proposed regulation therefore incorporates Census division differences
in historical HPI values and historical measures of the dispersion of
house values around levels suggested by the HPI. See section
III.A.4.d., Property Valuation.
NAR did not recommend a specific approach, but cautioned that an
REO price index might not be meaningful for Enterprise loans, because
the Enterprises tend to sell REO properties quickly, thus limiting
exposure to undue loss of value. For that reason, NAR recommended that
any analysis of REO property values be based solely on Enterprise data.
OFHEO also concurs with NAR that an REO price index built on non-
Enterprise data might be of limited usefulness for Enterprise loans.
Given the richness and volume of the Enterprise data, and consistent
with all other parts of the stress test, OFHEO has based the model of
REO property values on Enterprise data. However, rather than developing
a separate price index for REO properties, the proposed stress test
models REO property value as a function of the path of the HPI. In
addition, OFHEO proposes to adjust the resulting rate of loss of
principal balance rate to reflect the fact that REO property values in
the benchmark loss experience were lower in relation to the HPI than
the REO property values in other Enterprise experience.
d. Multifamily Loss Severity
With respect to loss severity, the stress test uses the same cost
elements for multifamily loans as for single family loans. However,
there is no loan seasoning, nor is statistical analysis used to
determine loss of loan principal balance. All cost and revenue elements
of multifamily loss severity rates are averages from Enterprise
experience.
7. Relating Losses to the Benchmark Loss Experience
The 1992 Act specifies that the stress test should apply rates of
default and loss severity that are ``reasonably related'' to the
highest rates experienced by the Enterprises for a period of at least
two years in any contiguous areas having at least five percent of the
nation's population (the benchmark loss experience).\129\ The stress
test satisfies this reasonable relationship requirement in the context
of two severe interest rate environments that are quite different from
the interest rate environment of the benchmark loss experience. At the
same time, the stress test also accounts for appropriate distinctions
in credit risk across loan types and characteristics. OFHEO believes
that the multivariate mortgage performance models developed by OFHEO
are the best means of specifying loss rates for the wide variety of
loans held by the Enterprises under the different interest rate
scenarios specified in the statute. However, for reasons explained
below, the models are adjusted to produce loss rates that are
reasonably related to the losses experienced on the 30-year fixed-rate,
single family mortgages in the benchmark time and place.
---------------------------------------------------------------------------
\129\ 1992 Act, section 1361(a)(1) (12 U.S.C. 4611(a)(1)).
---------------------------------------------------------------------------
Both Fannie Mae and Freddie Mac provided comments on how to
implement a statistical model of mortgage performance that would be
reasonably related to the benchmark loss experience. As discussed
earlier, neither Fannie Mae nor Freddie Mac recommended a joint,
multivariate statistical model of conditional default and prepayment
rates. However, both discussed how other models could be used in the
stress test and commented that a reasonable relation to the benchmark
loss experience could be achieved by estimating those models solely on
data from the benchmark loss experience.\130\ They noted that the
advantage of limiting the statistical sample in that way is to allow
the resulting equations to capture benchmark economic conditions
without having explicit explanatory variables for economic conditions
in the stress test.
---------------------------------------------------------------------------
\130\ Fannie Mae recommended estimation of a statistical model
of total terminations and Freddie Mac recommended estimation of a
statistical model of prepayments only.
---------------------------------------------------------------------------
The suggestion from Fannie Mae and Freddie Mac that the mortgage
performance models be estimated solely with data from the benchmark
loss experience, although appealing conceptually, turned out to be
impractical. The benchmark loans comprise too small and homogeneous a
set of loans to estimate models for all the Enterprises' current loans.
Using a much larger sample of historical loan performance experience
was important when estimating the statistical models, because it
provided a wide variety of economic circumstances and mortgage
experience upon which to base estimation of the model parameters. Like
current Enterprise loan portfolios, the samples used to estimate the
statistical equations include mortgages originated over many years and
geographic locations, and having distributions across other factors of
mortgage performance--such as age, coupon type or amortization terms--
that differ from those of the benchmark loans.
The ``reasonable relationship'' requirement of the 1992 Act means
that the adverse credit stress of the benchmark loss experience should
be reflected in the stress test mortgage losses. However, when the
mortgage performance models are applied unadjusted to a pool of loans
with the same characteristics as the benchmark loans, using interest
rate and house-price appreciation paths equivalent to those of the
benchmark time and place, the resulting default and severity rates are
slightly lower than the actual rates for the benchmark loss experience.
This result should be expected, because the mortgage performance models
are estimated from data on a broad range of historical experience,
rather than just data from the benchmark loss experience. The benchmark
loss experience was from the time and place with the worst mortgage
losses for the Enterprises. Therefore it is reasonable to expect it to
have default and severity rates somewhat higher than would be predicted
based solely upon the explanatory variables used in the stress test.
For this reason, the stress test
[[Page 18143]]
includes adjustments to the models to reflect more fully the additional
stress of the benchmark experience.
OFHEO proposes to relate losses projected by the statistical
equations to the benchmark loss experience in two ways. First,
benchmark house-price growth rates and multifamily (rental) market
economic conditions that coincide with the time and place of the
benchmark loss experience are applied to loans in the starting
portfolio during the stress test period. Second, the default and
severity rates predicted by statistical equations are increased, or
``calibrated,'' to the benchmark loss experience rates, so that if
newly originated loans with similar characteristics to those comprising
the benchmark sample were subjected to the same economic circumstances
as occurred in the benchmark loss experience, the statistical model of
mortgage performance would project ten-year cumulative default and
average severity rates equal to the rates actually observed for the
benchmark sample.\131\ Under this approach, default and loss severity
rates differ from the benchmark rates only to the extent interest
rates, property values, and loan characteristics are different from the
benchmark sample, or to the extent adjustments are necessary to account
for other statutory requirements.\132\ Because of the addition of this
benchmark ``calibration'' factor to default and loss severity
equations, loss rates for all loans are slightly higher than would
otherwise be projected.
---------------------------------------------------------------------------
\131\ Loans comprising the benchmark sample were 30-year fixed-
rate loans.
\132\ Differences in interest rates, property values, and loan
characteristics can have very significant effects, however. The
average mortgage credit loss rate for the two Enterprises in the
benchmark sample was 9.4 percent. In the up-rate scenario of the
stress test for June 1997, the average loss rate was 1.8 percent,
while in the down-rate scenario it was 1.4 percent. The loss rate
for the benchmark sample does not take account of mortgage insurance
and other credit enhancements. Losses on benchmark loans after
accounting for these receipts would have been seven percent.
---------------------------------------------------------------------------
Although the principles for reasonably relating stress test losses
to the benchmark loss experience are the same for single family and
multifamily loans, the methods of reasonably relating losses to the
benchmark differ and are discussed separately below.
a. Single Family Calibration
For single family loans, calibration constants are added to default
and loss severity rates.\133\ These constants are set forth in sections
3.5.2.3.2.9 and 3.5.3.3.3 of the Regulation Appendix. Their development
is described in section IV.B.8., Consistency with the Historical
Benchmark Experience, of the Technical Supplement.
---------------------------------------------------------------------------
\133\ The calibration constant used in the single family default
rate equations is in addition to the particular product-type
multiplier factors discussed earlier. The product-type multipliers
relate other products to the benchmark 30-year fixed-rate loans,
while the calibration constant relates all loans to the severe
benchmark loss experience.
---------------------------------------------------------------------------
The calibration constants were computed in three steps. First, all
benchmark loans were assigned the same historical house-price
experience--the ten-year sequence of appreciation rates from the OFHEO
HPI for the West South Central Census Division, commencing in 1984,
first quarter.\134\ Second, using the statistical equations estimated
on a broader historical loan sample, OFHEO projected the ten-year
experience of loans comprising the benchmark sample, computing the ten-
year cumulative default rate and ten-year average loss severity rate.
These rates were measured in the same manner for the benchmark in
NPR1.\135\ Third, these cumulative rates were compared to the actual
cumulative default and prepayment rates computed for the benchmark in
NPR1, and adjustment constants were calculated that, when applied in
the models, would yield the equivalent default and loss severity rates.
---------------------------------------------------------------------------
\134\ The West South Central Census Division does not exactly
match the four-State benchmark region, but its use here to represent
benchmark economics is consistent with OFHEO's proposal to aggregate
data based on Census divisions and to apply historical Census
division-level house price growth rates to season loans at the
beginning of the stress test. What is most important is that the
price series used to calibrate the statistical equations is the same
series that will be used in the stress test itself. The actual ten-
year house-price experience of the West South Central Division and
the four-State benchmark area, 1984-1993, are very similar.
\135\ The ten-year cumulative default rate was computed as the
sum of original UPBs for defaulted loans, divided by the sum of
original UPBs for all loans in the sample. The average severity rate
was calculated in similar fashion. Following the method used to
identify the benchmark experience, the calibration procedure
computes ten-year default and severity rates for each Enterprise
separately, and then the two Enterprise-specific rates are averaged.
---------------------------------------------------------------------------
The adjustment constant for loss severity rates is not applied to
the entire loss severity rate, but rather to the loss of loan principal
balance element of the loss severity rate. The constant is computed by
subtracting the loss of loan principal balance that was predicted by
the single family loss severity model from the loss of loan principal
balance that occurred on defaulted loans in the benchmark loss
experience. The second element of severity cost, transaction costs, was
not adjusted to reflect benchmark conditions. OFHEO found it more
appropriate in a national stress test to use a national blend of the
institutional factors such as foreclosure costs, property management
fees, and property sales expenses that comprise this element. The third
element of loss severity cost, asset funding costs, enters the stress
test as an imputed interest cost. As described in more detail in
section 3.5.3 of the Regulation Appendix, this element is related to
the benchmark loss experience through the use of foreclosure and
property disposition event timing from the benchmark loss experience.
The timing of these events determines the periods over which funding
costs are calculated.
b. Relating Other Single Family Products to the Benchmark
In the ANPR, OFHEO asked how to relate other types of mortgages to
the benchmark, which was developed based on single family, 30-year,
fixed-rate mortgages. The commenters' consensus was that some type of
multiplier approach to alternative single family mortgages should be
used, except for ARMs. These comments are discussed below.
(i) ANPR Comments
NAR suggested that OFHEO develop statistical models of default for
fixed- and adjustable-rate mortgages and relate the performance of
other mortgage types to them. NAR also pointed out, however, that this
type of relationship might be difficult to establish for new mortgage
types for which there is insufficient historical experience. NAR
suggested applying the benchmark default experience to these loans
rather than measuring the difference in risk from the benchmark
experience. VA addressed the same concern, suggesting that multipliers
should be based on historical periods in which the other mortgage types
had significant shares of the market. Specifically, VA suggested that
measures of performance from those periods of other single family
mortgage types relative to the 30-year, fixed-rate product could be
used to impute the necessary performance differences from the benchmark
loss experience to use in the stress test. Freddie Mac stated that any
default-rate multipliers should be based on a broader range of
Enterprise historical experience than the benchmark time and place.
Freddie Mac, although recommending that OFHEO use simple
multipliers, also raised a concern that loans receiving multiple
multiplier factors could end up with unreasonably high stress test
default rates. It cited, as an example, a balloon loan on an investor-
owned condominium. If the stress test were to apply default-rate
multipliers for each of these three mortgage type categories
[[Page 18144]]
(condominium, investor-owned, and balloon), the combined risk factor
premium could be unreasonably high. To remedy this problem, Freddie Mac
recommended that the stress test incorporate limits on the interaction
of risk factors.
MRAC suggested that, if sufficient data were available, OFHEO might
either create historical tables of default rates by various loan
characteristics, in order to establish product-type multipliers, or use
some type of regression analysis to discern performance differences
among mortgage types. The MBA suggested that multipliers are the best
approach because they are currently used by the Enterprises and
therefore would provide a simple way for them to implement the risk-
based capital standards.
OTS cautioned that multipliers might not be appropriate for ARMs or
for multifamily loans, because the credit loss experience of these
loans may not correlate well with that of fixed-rate, single family
loans. OTS recommended that OFHEO consider using separate benchmarks
for different types of loans. ACB, however, commented that there is no
statutory requirement to incorporate the worst experience for each
mortgage type into the stress test, and that a multiplier analysis for
single family loan types would be sufficient.
Consistent with its recommendation that OFHEO not develop a
statistical model of conditional default rates, Fannie Mae suggested
that multipliers be applied to (cumulative) loss rates, rather than to
conditional default rates.
(ii) OFHEO's Response
The stress test approach of adding product type adjustment factors
as explanatory variables in a single family default equation is
consistent with the multiplier approach recommended by commenters.
However, the stress test approach does not have the shortcomings about
which some commenters cautioned. It relies upon a broader historical
experience than the benchmark sample alone to gauge the relative risk
of other mortgage types, and it controls for the multiple multipliers
problem outlined by Freddie Mac. The multiple multipliers problem is
avoided because product type adjustment factors are estimated as part
of the statistical default equation. The equation computes the marginal
impact of each product type after controlling for all other explanatory
variables. Using simple multipliers with limits on the amount of
adjustment, as recommended by Freddie Mac, would either be too
imprecise to reflect the relative risk of the loans that fall into
multiple product type categories, or else would become as complex as a
statistical model in order to account for all of the conceivable
combinations of product types.
OFHEO agrees with the OTS comment that a multiplier approach is not
appropriate for ARMs. Equations for single family default and
prepayment rates in the stress test are, therefore, estimated
separately for ARMs. This is appropriate because the adjustable payment
features of these loans create unique incentives to either default or
prepay that are not found in other mortgage types. The ARM default
equation does, however, receive the same benchmark calibration constant
used in the other two single family default equations. The use of this
constant reasonably relates ARMs to the added stress of the benchmark
loss experience in a manner consistent with how other single family
product types are related to the benchmark loss experience.
c. Relating Multifamily Mortgage Performance to the Benchmark
In the ANPR, OFHEO requested comment on how the stress test
multifamily mortgage performance should be related to the single family
benchmark. Respondents to the ANPR mentioned the need to capture the
different underwriting variables and economic factors that would
influence multifamily performance directly. They warned against
applying multipliers to single family losses to generate multifamily
losses. These concerns were raised by OTS, MBA, Fannie Mae, and Freddie
Mac. In addition, OTS and Fannie Mae suggested that OFHEO may need to
explore options other than relating stress test credit losses on
multifamily loans to the single family benchmark.
OFHEO agrees with the commenters' concerns about using a simple
multiplier approach for multifamily loans, and proposes instead a
separate statistical model of multifamily mortgage performance based on
multifamily market conditions, property financial characteristics (DCR
and LTV), and loan terms--whether fully amortizing or balloon, or
having fixed or adjustable interest rates. The statistical model allows
the application of OFHEO's first principle, outlined above in section
III. A. 5. e., Choice of Explanatory Variables for Default and
Prepayment, for relating stress test losses to the benchmark: using
economic conditions of the benchmark experience in the stress test.
OFHEO believes that multifamily rent and vacancy indexes from the
benchmark time and place provide the best means to relate starting
multifamily loan portfolios to the benchmark loss experience. These
indexes account for the economic decline that occurred in the benchmark
region in the economic factors that affect multifamily mortgage credit
risk. Therefore, the stress test creates a reasonable relationship to
the benchmark loss experience by using vacancy rates from and percent
changes in rents from the benchmark loss experience to update property
financials (DCR and LTV) throughout the stress period.
Because of the small number (13) of multifamily loans purchased by
the Enterprises in the benchmark region during 1983 and 1984, it is not
possible to compute calibration adjustments like those in the single
family default and severity equations. Instead, OFHEO proposes to treat
all defaults as full foreclosure events and apply loss severity rates
without consideration of loan seasoning. The effect of this approach is
to create higher credit losses than if the stress test were to account
for multifamily defaults that are resolved without foreclosure and
adjust severity rates to account for the age of loans.
Methodologically, treating all multifamily defaults as foreclosure
events is consistent with OFHEO's proposed approach to single family
credit loss generation in the stress test. However, OFHEO is aware that
use of various default resolution strategies other than foreclosure
(loss mitigation) played an important role in controlling multifamily
default losses in the severe environment of the late 1980s and early
1990s. Therefore, accounting for loss mitigation in the stress test
would tend to decrease losses for any given economic conditions.
Treating all defaults as foreclosures for calibration purposes, rather
than allowing for loss mitigation efforts, results in an increase in
loss severity--before application of any credit enhancements--of 6.5
percent per defaulting loan.\136\
---------------------------------------------------------------------------
\136\ The 6.5 percent figure is arrived at by multiplying the 13
percent of defaults resolved with alternatives to foreclosure by a
50 percent loss rate reduction factor.
---------------------------------------------------------------------------
There is an exception to the rule of treating all defaults as
foreclosure events for Enterprise loan programs that require the
seller/servicer to repurchase loans that become 90-days delinquent. For
loans in these programs, the recorded ``default'' event at the
Enterprises is the point at which a loan becomes 90 days delinquent,
rather than a foreclosure-like event where the Enterprise obtains title
to the collateral property.
[[Page 18145]]
The stress test loss severity rate for these loans is 39
percent.\137\ The 39 percent loss severity rate reflects experience of
the Enterprises during the stressful conditions of the early 1990s,
including approximately 50 percent cures (or modifications) and 50
percent foreclosures on 90-day delinquencies. OFHEO research indicates
that this is a reasonable approximation for the stress test.
---------------------------------------------------------------------------
\137\ This rate is discounted by 12 months to reflect the
average time from the default date (30 days after last paid
installment date) to final resolution.
---------------------------------------------------------------------------
8. Inflation Adjustment
The 1992 Act specifies that, to the extent that the ten-year CMT
increases by more than 50 percent over its average for the nine months
preceding the starting date of the stress test, credit losses must be
adjusted ``to reflect a correspondingly higher rate of general price
inflation.'' \138\ In the stress test, mortgage credit losses are not
related to rates of general price inflation, but most are related to
rates of house price inflation.\139\ Implementing this provision of the
statute requires consideration of the relationship between interest
rates, general inflation rates, and house price inflation rates.
---------------------------------------------------------------------------
\138\ 1992 Act, section 1361(a)(2)(E) (12 U.S.C. 4611(a)(2)(E)).
\139\ Multifamily credit losses are related to rent growth
rates. The same adjustment described here for house price inflation
rates is also made to rent inflation rates.
---------------------------------------------------------------------------
These relationships are complex. Over recent decades, changes in
broad inflation measures generally have preceded changes in interest
rates in the same direction. And changes in interest rates have been
accompanied by changes in house price inflation rates in the opposite
direction. Thus, over short and intermediate periods of time, interest
rates and house price inflation rates have often moved divergently. For
example, consider the three five-year periods beginning in 1975. From
the beginning of 1975 to the end of 1979, the ten-year CMT averaged
about 8 percent, while house prices rose at an 11 percent annual rate.
In the following five-year period, from 1980 to 1984, interest rates
were 50 percent higher (12 percent), while house price inflation fell
to 4 percent. Then in the third five-year period, 1985 to 1989,
interest rates declined to 9 percent, while house price gains
accelerated to 7 percent.\140\ Over longer periods of time, however,
these changes have tended to reverse themselves. For periods of ten
years or more, higher (lower) than average interest rate levels have
generally been associated with higher (lower) than average rates of
general inflation and house price inflation.
---------------------------------------------------------------------------
\140\ General inflation rates (based on the CPI) followed a
still different pattern. They averaged 8 percent per year during the
first five-year period, 7 percent in the second, and 3 percent in
the third five-year period.
---------------------------------------------------------------------------
In unusual environments, such as those represented by the economic
conditions of the stress test, average past relationships between
interest rates, general inflation rates, and house price inflation
rates may not prevail. The nature or cause of the projected mortgage
credit stresses in the stress test are not specified in the statute.
They could involve problems particular to housing markets, such that
house price behavior deviates persistently from general inflation
patterns. Or they could be focused on non-house-price factors, such as
unemployment, relocation, or divorce rates.
Except to the extent that the ten-year CMT rises in the up-rate
scenario by more than 50 percent, the stress test does not project any
differences in house price changes or other sources of credit stress in
the two interest rate scenarios. And, aside from the inflation
adjustment, the specific pattern of house price changes used in both
scenarios is not designed to be consistent with any particular pattern
of interest rates. It was chosen to replicate (and encapsulate in one
variable) the overall level of credit stress in the benchmark loss
experience.
In order to implement the statutory requirement, the stress test
projects that cumulative increases in house prices, a component of
general inflation, are higher in the up-rate scenario by an amount that
reflects, percentage point for percentage point, any positive
difference between the ten-year CMT and the level corresponding to a 50
percent increase. Thus, for example, if the ten-year CMT starts at 6
percent and increases by 75 percent to 10.5 percent, the increase in
excess of 50 percent is 1.5 percentage points. The cumulative change in
house prices during the up-rate scenario would equal the cumulative
change during the down-rate scenario plus an upward adjustment. The
adjustment is the amount needed to reflect what the cumulative increase
would be if the house price inflation rate were 1.5 percent higher, on
average, throughout the part of the stress period in which the ten-year
CMT exceeds 9 percent.\141\
---------------------------------------------------------------------------
\141\ The stress test would calculate the cumulative adjustment
factor in this case to be 1.0159\1/6\, so final house
price levels in the up-rate scenario would be 14.6 percent higher
than they would be in the down-rate scenario. In this formula, 9\1/
6\ represents the number of years the ten-year CMT exceeds 9 percent
by the full 1.5 percentage points plus two months to reflect the
period in which the ten-year CMT exceeds 9 percent by a smaller
amount. If the ten-year CMT increases 75 percent over the base
month, a 50 percent increase will be achieved by month eight. The
full increase will be achieved by month 12. For the purposes of this
calculation, the result is the same as it would be if the extra 25
percent lasted for nine years and two months.
---------------------------------------------------------------------------
In recognition of the likely short- and intermediate-term
divergence between interest rates and house price behavior, the stress
test concentrates all of the adjustment in the final five years of the
stress period. Thus, house prices are identical in the two stress test
interest rate scenarios during the first five years, but increase much
more rapidly in the last five years of the up-rate scenario than they
do in the down-rate scenario.
Several respondents to OFHEO's ANPR commented on this issue. VA
opposed any adjustment, arguing that while the long-term behavior of
house price inflation and general inflation is consistent, the short-
term relationship is weak, and the relationship between interest rates
and house prices ``is even more tenuous.'' VA further agrees that
specific economic conditions can disrupt any general relationships, and
that an adjustment would be inconsistent with the approach of private
rating agencies. OFHEO believes, however, that some adjustment is
required by the statutory language.
HUD argued that adjusting the rate of increase in house prices
throughout the stress period on a one-to-one basis with general price
inflation would deny the role of changes in real interest rates over
time. HUD suggested that OFHEO consider current trends and long-run
relationships between real interest rates and house prices. NAR
suggested that a one-to-one relationship is appropriate for long-term
assumptions, and ACB commented similarly. OFHEO believes that its
approach, which uses a one-to-one relationship for the cumulative
change but concentrates the change in the last five years of the stress
period, is not inconsistent with any of these recommendations.
Freddie Mac recommended that house price inflation should vary with
interest rates in a one-to-one relationship, not only with respect to
increases in the ten-year CMT exceeding 50 percent, but also with
respect to all interest rate changes. House price inflation rates would
be based on rates current at the start of the stress period and rise or
fall by amounts equal to the change in the ten-year CMT in both
scenarios. Such an approach could result in more severe credit losses
in the down-rate scenario and very few credit losses in the up-rate
scenario. OFHEO believes that the stress test should reflect the
possibility that substantial credit losses would occur in either
scenario. The recommended
[[Page 18146]]
approach also would not have any obvious relationship to the benchmark
loss experience. Applying the approach at the time the benchmark loans
were originated would result in much stronger house price growth than
actually occurred in the benchmark area.
Freddie Mac further argued that a stress test that incorporated a
ten-year CMT that exceeded the rate of house price appreciation by more
than 6.5 percentage points over a ten-year period would be inconsistent
with national historical experience and, therefore, inappropriate.
However, national historical experience is not an appropriate criterion
for the stress test's key source of mortgage credit stress. Credit
losses in the stress test are required to exceed national historical
experience. They are based on the worst regional, not national,
experience.\142\ More importantly, as discussed above, house price
projections in the stress test are not designed to correspond to any
particular interest rate level. Rather, they are simply a means of
incorporating an overall credit stress level that is comparable to the
benchmark loss experience and which may reflect stresses from a variety
of non-house price sources not explicitly included in the mortgage
performance model.
---------------------------------------------------------------------------
\142\ The average ten-year CMT exceeded average house price
growth in the West South Central Division during the 1980s by 9.5
percentage points. For the benchmark loss experience, the difference
was 8.5 percentage points.
---------------------------------------------------------------------------
B. Interest Rates
The 1992 Act specifies the level of the constant maturity Treasury
yield (CMT) for ten-year securities during the last nine years of the
stress period.\143\ However, only general guidance is provided for the
levels of yields on Treasury securities with different maturities.
Also, yields on other financial instruments are not explicitly
mentioned. The behavior of yields on financial instruments other than
ten-year Treasury securities will have potentially substantial and
pervasive effects on the Enterprises during the stress period. Those
yields will determine the cost of new debt issued and earnings on new
investments, as well as the interest rates paid or earned on assets,
liabilities, or derivatives contracts that are tied to market yield
indexes. They will also have a significant effect on the volumes of
mortgage prepayments and defaults. The magnitude of the effects on an
Enterprise during the stress period will depend greatly on the
Enterprise's funding strategies at the start of the stress period.
---------------------------------------------------------------------------
\143\ 1992 Act, section 1361(a)(2) (12 U.S.C 4611(a)(2)).
---------------------------------------------------------------------------
1. Yields on Treasury Securities
a. Statutory Requirements
The 1992 Act describes two interest rate scenarios (one rising and
one falling) based on movements in the ten-year CMT. In the rising or
up-rate scenario, the ten-year CMT increases during the first year of
the stress test period and then remains constant at the greater of: (1)
600 basis points above the average yield during the preceding nine
months; or (2) 160 percent of the average yield during the preceding
three years. However, in no case may the yield increase to more than
175 percent of the average yield over the preceding nine months. In the
falling or down-rate scenario, the ten-year CMT decreases during the
first year of the stress period and then remains constant at the lesser
of: (1) 600 basis points below the average yield during the preceding
nine months; or (2) 60 percent of the average yield during the
preceding three years. However, in no case may the yield decrease to
less than 50 percent of the average yield over the preceding nine
months.
The 1992 Act does not specify the shape of the yield curve during
the stress period. Rather, it simply requires that the levels of other
Treasury yields ``change relative to the 10-year Constant Maturity
Treasury (CMT) yield in patterns and for durations that are reasonably
related to historical experience and are judged reasonable by the
Director.'' \144\ The statute also does not specify the manner in which
the ten-year CMT moves during the first year of the stress period to
reach the level required for the remainder of the period.
---------------------------------------------------------------------------
\144\ 1992 Act, section 1361(a)(2)(D) (12 U.S.C. 4611(a)(2)(D)).
---------------------------------------------------------------------------
In its comments to OFHEO's ANPR, ACB suggested that OFHEO consider
using stochastic projections of all interest rates, if OFHEO determined
that stochastic projections were consistent with statutory
requirements. ACB noted that the process could be constrained to insure
that the ten-year CMT reached its required level during the final nine
years of the stress period on an average basis. OFHEO has determined
that such an approach would not be compatible with the 1992 Act. That
statute clearly specifies that the ten-year CMT will be constant during
the final nine years of the stress period. Furthermore, as Fannie Mae
commented, using a stochastic model for determining interest rates
would create unnecessary uncertainty about what amount of capital would
actually be required for a given set of risk positions. A stochastic
model also would add unnecessary complexity to the regulation.
Accordingly, OFHEO proposes that all interest rates during the stress
period be fully determined by past data on interest rates.
b. Yields of Other Treasury Maturities During the Final Nine Years
(i) Constant or Varying Yields
OFHEO considered whether the Treasury yield curve should be
constant over the final nine years of the stress period or whether it
should change in some specific manner. OFHEO proposes to use a constant
yield curve. While yields are extremely unlikely to remain constant or
even roughly so over a period as long as nine years, there are no
serious disadvantages to using such an approach in the stress test, and
there are compelling advantages.
A constant yield curve is a straightforward approach that is
consistent with the statutory specification of a constant ten-year CMT.
The purpose of the interest rate component of the stress test is to
assess an Enterprise's ability to withstand a prolonged shift to a much
higher or much lower interest rate environment. No specific pattern of
yield changes can fully capture the range of possible future adverse
changes. Based on historical experience, one would expect all interest
rates to fluctuate over a broad range during a period as long as nine
years. Different underlying macroeconomic circumstances would be
associated with different evolutions of the entire yield curve,
including the ten-year CMT. Tying the stress test to one specific set
of macroeconomic circumstances would tend to limit its general
usefulness. The real-life danger the Enterprises face of much higher or
much lower interest rates during the next decade is not focused on any
particular portion of that ten-year period. Designing a stress test
with any specific pattern of interest rate changes after the first year
of the stress period would imply a belief that Enterprise risk
exposures in some future years would be a matter of greater public
concern than in other years. While an argument could be made that near-
term risk exposures would create losses with a higher present value,
that concern should be balanced by a recognition that the risk of a
very different interest rate environment is greater for distant years
than for the near-term.
A stress test with interest rates that are especially high or low
in particular
[[Page 18147]]
future years would encourage Enterprise hedging strategies to focus on
those specific years. Risks in other years, when stress test
projections were more moderate, might receive relative neglect. The
Enterprise would thus be providing more protection against more
adverse, but less likely, interest rates in some years at the expense
of less protection against less adverse, but more likely, interest
rates in other years. Such an incentive would provide less general
protection and thereby increase the risk of failure.
In their ANPR comments, Fannie Mae and VA suggested specific fixed
yield curves, consistent with OFHEO's proposal in this regard. Freddie
Mac recommended a considerably more complex approach that would
generally result in relatively more adverse short-term interest rates
in the early part of the final nine years of the stress period and less
adverse short-term interest rates later. OFHEO believes its proposal is
much simpler and will provide better general protection against
Enterprise failure for the reasons discussed above.
Freddie Mac argued that a fixed yield curve would be unreasonable
for two reasons. First, Freddie Mac stated that a fixed curve would be
inconsistent with the statutory requirements that changes in yields on
Treasury securities with maturities other than ten-years ``will change
relative to the 10-year constant maturity Treasury yield in patterns
and for durations that are reasonably related to historical
experience.'' It is clear from the legislative history that Congress
did not intend to prohibit constant yield curves, per se, but rather
wanted to prohibit unusual yield curves lasting for a longer time than
could be reasonably related to historical experience. The language of
the statute follows the original Senate-passed bill, except that
``reasonably related to'' in the quoted phrase was substituted for
``within the range of,'' and a specific restriction on unusual yield
curves was removed. The Senate Committee, in explaining its
understanding of the yield curve provision, actually recommended that
the yield curve be fixed during at least the final five years of the
stress period.\145\
---------------------------------------------------------------------------
\145\ S. Rep. No. 102-282, at 22 (1992).
---------------------------------------------------------------------------
Second, Freddie Mac argued that a constant yield curve ``would be
of little value in measuring the ability of an Enterprise to absorb
losses in relation to its risks'' because interest rate volatility
would disappear and the prices of options would approach zero. Market
estimates of interest rate volatility, however, play no important role
in the stress test OFHEO is proposing. The Enterprises are not
projected to buy or sell any options, as this is a ``no new business''
stress test. While option value does affect decisions about option
exercise, and those decisions are an important element of the stress
test, the interest rate movements in the stress test are quite large.
In such circumstances, Enterprise decisions about option exercise will
generally be relatively insensitive to precise measures of option
value. Homeowners' decisions to exercise their options to prepay their
mortgages are also based on past homeowner responses to large changes
in interest rates and not on specific measures of volatility. Stress
test projections relating to the exercise of options implicitly assume
that expectations about volatility are within normal ranges, despite
the lack of change in interest rates. The proposed approach is an
efficient simplification that does not distort Enterprise risks in any
meaningful way.
(ii) Choice of Fixed Yield Curve Shapes
OFHEO proposes that all Treasury yields for key maturities (three-
and six-month; one-, three-, five-, and 20-year) in the final nine
years of the up-rate scenario be equal to the ten-year CMT. In the
final nine years of the down-rate scenario, OFHEO proposes that all key
Treasury yields have the same ratio to the ten-year CMT that they had,
on average, during the nine-year period from May 1986 through April
1995. The proposed yield curves for both interest rate scenarios
correspond to historical experience.
OFHEO based its selection of yield curves on an examination of
historical data on Treasury yields. Data are available starting in
December 1958. OFHEO focused on the relationship between a short-term
(six-month) yield and the ten-year yield.\146\ From 1959 through 1996,
the average yield curve slope, measured by the ratio of the six-month
CMT to the ten-year CMT, was 0.88, a moderate upward slope. However,
when calculated on a monthly basis, this slope has varied considerably
through time (See Table 26, Frequency Distribution of Yield Curve
Slopes, 1959--1996). Monthly slopes have been as low as 0.48 (September
and October 1992) and as high as 1.29 (March 1980). In more than half
of the months, yield curves were roughly flat or downward sloping
(slopes above 0.95) or were steeply upward sloping (slopes below 0.75).
---------------------------------------------------------------------------
\146\ In the following discussion, yields of six-month Treasury
bills are expressed on a bond-equivalent basis. The six-month
maturity has the advantage that the timing of its payments are
consistent with the interest rate payment cycle of Treasury notes
and bonds, ensuring comparability of yields across maturities.
[[Page 18148]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.211
Of particular relevance are the average slopes over periods of 108
months (nine years) and their relationship to previous increases or
decreases in yields. Ratios of the average six-month Treasury CMT to
the average ten-year CMT for periods of 108 months ranged from 0.77
(for periods ending from January 1994 through April 1996) to 0.99 (for
periods ending from September 1981 through June 1982). OFHEO must
project yields curves for a nine-year period in which the ten-year CMT
has increased by 75 percent, and decreased by 50 percent, from its
average in the nine months ending one year before the beginning of the
nine-year period.\147\ Accordingly, OFHEO sought to determine whether
historical data suggest any relationship between changes in average
ten-year CMT and yield curve slopes for relevant time periods.
---------------------------------------------------------------------------
\147\ In high yield environments, the changes in interest rates
would be somewhat smaller, but past and recent data suggest that the
changes will generally be of this magnitude.
---------------------------------------------------------------------------
At no time during the past 40 years have ten-year CMTs changed as
greatly as required in the stress test. The largest comparable increase
was 56.3 percent from the nine-month average of 6.04 percent during
November 1971 to July 1972 to the nine-year average of 9.44 percent
during August 1973 to July 1982. The ratio of six-month to ten-year
yields during the later period was 0.98. The largest comparable
decrease was 38.9 percent from the nine-month average of 12.74 percent
during February to October 1984 to the nine-year average of 7.78
percent during November 1985 to October 1994. That change was
associated with a slope of 0.77 during the nine-year period.
The pattern of relatively flat yield curve slopes after interest
rate increases and steep yield curve slopes after interest rate
decreases is consistent with the data. In all nine-year periods in
which the average ten-year CMT was above its average during the
relevant earlier nine-month period, the yield curve slope was greater
than 0.87. In all nine-year periods in which the average ten-year CMT
was below its average during the relevant earlier nine-month period,
the yield curve slope was less than 0.87. Furthermore, the greater the
increase in the ten-year CMT, the flatter the yield curve slope tended
to be, and the greater the decrease in the ten-year CMT, the steeper
the yield curve slope tended to be. Results of an ordinary least
squares regression imply that a sustained 75 percent increase in the
ten-year CMT would likely result in a CMT yield curve slope of 1.00,
while a sustained 50 percent decline provides an expected slope of
0.77.\148\
---------------------------------------------------------------------------
\148\ An ordinary least squares regression describes the results
quantitatively. The dependent variable (Yt) is the ratio
of the average six-month CMT to the average ten-year CMT during the
nine years ending in month t. The independent variable
(Xt) is defined as the ratio of the average ten-year CMT
in the nine years ending in month t to the nine-month average of the
ten-year CMT from month t-128 to month t-120. The regression results
are: Yt = 0.86 + 0.19 Xt.
Although this regression is based on monthly data over a 38-year
period, it is a small data set for investigating this issue. The
yield data start in December 1958, but each observation needs 128
months prior data, so the first observation used in the regression
is August 1969. That leaves 326 observations through September 1996,
but because of the lags, each observation is very similar to the one
preceding it. There are really only four fully separate dependent
variable observations. In these circumstances, the coefficient
estimates are unbiased, but the usual regression statistics are not
meaningful. In an alternative regression, the data were reorganized
as follows. The 326 observations were rank-ordered by the
independent variable and divided into quartiles. Using average
values of the two variables from each quartile, the regression was
rerun with the resulting four observations. The results are:
Yt = 0.86 + 0.20 Xt.
Differences in parameter estimates from the full sample
regression were small, less than 0.01, and the standard error of the
coefficient of Xt was 0.022. Even though the observations for these
regressions were limited, to the extent the data do exist, they
support OFHEO's yield curve proposal.
---------------------------------------------------------------------------
If the macroeconomic circumstances associated with a future shift
in yields were to differ from those that engendered interest rate
changes in recent decades, different results might easily occur.
Nevertheless, the historical experience of the past four decades, as
indicated both by the actual yield curve slopes in the episodes when
the ten-year CMT changed most greatly and by the more general results,
suggests an essentially flat yield curve in the up-rate scenario, and a
curve with a relatively steep upward slope in the down-rate scenario.
Although the highest yield curve slope was 0.99, OFHEO chose a more
straightforward yield curve slope of 1.00 for the up-rate scenario. The
largest historical interest rate increase resulted in an almost flat
yield curve, and that increase was still well below the increase of the
up-rate scenario of the stress test. In addition to the six-month
yields, OFHEO also proposes that all other key Treasury yields be equal
to the ten-year CMT in the up-rate scenario. When the six-month CMT
equals the ten-year CMT, setting all the other key
[[Page 18149]]
Treasury yields equal to the same levels is straightforward and
appropriate. In the down-rate scenario, however, setting the six-month
and the ten-year yields does not directly suggest appropriate rates for
instruments with other maturities. OFHEO proposes in this scenario that
slopes of key CMTs to the ten-year CMT be based on a specific
historical experience in a straightforward way that incorporates long-
term relationships between yields of instruments with different
maturities. The slope of the average six-month CMT to the average ten-
year CMT during the nine-year period ending in April 1995 closely
approximates the yield curve slope suggested by the regression
equation.
Several commenters responded to a question in OFHEO's ANPR about
the Treasury yield curve. Consistent with OFHEO's proposal, Fannie Mae
recommended that OFHEO focus its approach to projecting yield curves on
the ratio of the six-month Treasury yield to the ten-year Treasury
yield. However, Fannie Mae recommended that the ratio of the six-month
CMT to the ten-year CMT be set at a long-run historical average in both
interest rate scenarios. Such an approach would not be consistent with
actual experience that large sustained interest rate increases are
accompanied by relatively flat yield curves and that large, sustained
interest rate decreases are accompanied by relatively steep yield
curves.
The Department of Veterans Affairs recommended a yield curve
formula that would depend heavily on the shape of the yield curve at
the start of the stress test. OFHEO considered such an approach, but
found no evidence in historical data that the yield curve shape at the
start of a ten-year period is related to the average shape over the
final nine years of that period.
Freddie Mac suggested an approach based on an assumption that the
statutory changes in interest rates represent a ``regime shift.'' As
market participants adjust to the new regime, Freddie Mac argued,
average yield curve relationships should return. OFHEO believes it is
more appropriate to base projections of yield curve relationships on
what has actually occurred in the past with the most similar changes in
ten-year CMT levels.
NAR recommended that OFHEO take into account Treasury refunding
behavior during the stress period. In order to keep the stress test as
general as possible, OFHEO chose not to make any specific projections
about Treasury debt issuance during the stress period.
c. Yields of Treasury Securities During the First Year
OFHEO proposes that during the first year of the stress period, the
yields on Treasury securities of all maturities adjust linearly from
their levels in the month proceeding the stress period to their levels
during the final nine years of the stress period. In comments to
OFHEO's ANPR, Fannie Mae stated that movements of the six-month and
ten-year CMTs should be consistent during an adjustment period of one
to two years. OFHEO agrees and believes its proposal will result in
sufficiently consistent movement.
Freddie Mac suggested an approach under which, before the end of
the first year, the yield curve might invert in the up-rate scenario
and become very steeply upward sloping in the down-rate scenario. As
previously discussed, OFHEO believes this approach is unnecessarily
complex.
2. Yields of Non-Treasury Instruments
a. In General
Payments during the stress period associated with many Enterprise
assets, liabilities, and derivatives contracts and the performance of
mortgages, especially prepayment behavior, are dependent on future
levels of yields on non-Treasury instruments and levels of non-Treasury
interest rate indexes. OFHEO proposes to project these yield levels
using econometric models relating non-Treasury interest rate series to
yields on Treasury securities of comparable maturity.
The econometric specifications were based on two primary criteria.
First, whenever possible, the non-Treasury interest rate series were
modeled using the relative (rather than absolute) spread over
comparable CMTs. Second, the specifications balanced the desire for
simplicity with the need to account for the time-series properties
inherent in the data.
Autoregressive integrated moving average (ARIMA) models were used
to model the behavior of the non-Treasury interest rate series.\149\
The models capture the average historical relationships between
specific CMTs and non-Treasury interest rates. OFHEO believes this
approach is consistent with recommendations of all commenters to a
question on this issue in OFHEO's ANPR.
---------------------------------------------------------------------------
\149\ An ARIMA (p,d,q) model implies p autoregressive terms, d
differences of the original series, and q moving average terms.
Generally speaking, differencing is undertaken to render a series
``mean-stationary,'' which is a requirement for statistical analysis
of autoregressive models. For example, observations from a random
walk include the cumulative effect of all past shocks (random
disturbances) and/or trends. Differencing can net out the effect of
persistent movements and make a series stationary. Autoregressive
terms also represent the persistence of past shocks, but where the
effect of the shock diminished over time. Moving average terms
represent the effects of shocks that disappear completely after some
finite number of periods.
In some situations the original series may also exhibit non-
stationarity in the variance, requiring other normalizing
transformations (e.g., taking logarithms). Also, visual examination
of the data series and residual analysis based on appropriate
statistical criteria (e.g., Ljung-Box Q-statistics) were used to
guide the model selection process.
In some cases, a constant term has been included. This has the
effect of preserving the historical average relative spread between
the index and the corresponding Treasury rate when projecting future
values. This is only done when there is some evidence that this
historical difference is statistically significant. While
differencing is necessary in many models to achieve stationarity in
the mean, the use of relative spreads over Treasury rates of
comparable maturities generally appears to make the original
relative rate series variance stationary.
---------------------------------------------------------------------------
b. Yields on Enterprise Debt
OFHEO proposes that yields on Enterprise debt be projected in the
same manner as yields on other non-Treasury instruments, except that a
50 basis point premium is added after the first year of the stress
period. After one year of stress test conditions, the Enterprises might
appear strong based on accounting measures of earnings and net worth.
However, market values of the Enterprises' assets, liabilities, and
derivatives contracts would fully reflect the effects of the interest
rate shock and some of the credit quality deterioration of the stress
test. Investors would be aware of these changes in market value and
adjust their evaluations of the Enterprises' financial health
accordingly. Because the Enterprises' ability to withstand further
interest rate and credit shocks likely would be low, the Enterprises in
the final nine years of the stress period would likely not meet their
risk-based capital requirement and would, therefore, be subject to
dividend restrictions. Such events might strengthen investor concerns
about the Enterprises' financial health.
As government sponsored enterprises, the Enterprises likely would
suffer much smaller debt market penalties than fully private firms in
the same circumstances. However, the historical experiences of Fannie
Mae and the Farm Credit System during periods of financial stress
strongly suggest that borrowing costs would include some risk premium
during economic conditions such as those in the stress test. As
illustrated by data reported in the General Accounting Office's 1990
report on government sponsored enterprises, Fannie Mae's short-term
[[Page 18150]]
borrowing costs during 1980 through 1982 were generally about 80 basis
points in excess of yields on comparable maturity Treasury debt, rising
at one point to 200 basis points above Treasury yields. Spreads receded
after sharp declines in interest rates greatly improved Fannie Mae's
condition to a more normal range centered roughly at 20 basis points.
Spreads were high again in the late 1980s for both Fannie Mae and the
Farm Credit System, ranging from 40 to 100 basis points over a two-year
period during the Farm Credit System's time of greatest financial
difficulty.\150\
---------------------------------------------------------------------------
\150\ U.S. General Accounting Office (1990), Government
Sponsored Enterprises: The Government's Exposure to Risk,
Washington, DC: U.S. General Accounting Office, (GAO/GGD-90-97) 87-
88.
---------------------------------------------------------------------------
In stress test simulations based on the quarter ending in June
1997, the Enterprises' borrowing costs, including the 50 basis point
premium, are 78 basis points above comparable Treasury yields in the
up-rate scenario and 56 basis points above in the down-rate scenario
after the first year of the stress period. Such spreads are appropriate
because it is essential that the Enterprise be adequately prepared for
widening debt yield spreads in periods of financial stress.
In its comments to OFHEO's ANPR, ACB pointed to Fannie Mae's
difficulties in 1980 to 1982 as a possible basis for assessing likely
borrowing spreads in the stress period. ACB also suggested that OFHEO
might consider projecting the Treasury Department's use of its
statutory authority to lend money to the Enterprises in stressful
circumstances. OFHEO believes the stress test should assess the
Enterprises' abilities to withstand the stress test without borrowing
from the Treasury Department.
Freddie Mac commented that OFHEO should assume that the market's
perception of an implicit government guarantee on Enterprise debt
protects the Enterprises against any increased risk premium in
borrowing spreads. OFHEO disagrees and believes the historical evidence
is inconsistent with that view. OFHEO does agree that financial
weakness of the Enterprises during the stress period should not be
expected to have the same effect on borrowing costs that it would for
firms that are not government sponsored enterprises. Nonetheless, some
increase in risk premiums is appropriate. As the Enterprises' offering
prospectuses clearly state, Enterprise obligations are not backed by
the full faith and credit of the Federal government. OFHEO also agrees
that attempting to calculate appropriate borrowing spreads at different
times during the stress test, based on specific measures of Enterprise
stress, would unnecessarily complicate the test. Accordingly, OFHEO
proposes a constant risk premium during the final nine years of the
stress period.
C. Mortgage Credit Enhancements
1. Background
The Enterprises use mortgage credit enhancements to reduce their
credit risk exposure. For single family loans with LTV ratios in excess
of 80 percent, the Enterprises must use certain statutorily enumerated
credit enhancements. The Charter Acts prohibit the purchase of
conventional single family mortgages with LTV ratios in excess of 80
percent unless: (1) the seller retains a participation interest of 10
percent or more; (2) the seller agrees to repurchase or replace the
mortgage upon default (seller recourse); or (3) the amount of the
mortgage in excess of 80 percent is insured or guaranteed.\151\
Multifamily mortgages are not subject to such a requirement, but may
also be credit enhanced.
---------------------------------------------------------------------------
\151\ See sections 305(a)(2) and (4)(C) of the Federal Home Loan
Mortgage Corporation Act (12 U.S.C. 1454(a)(2) and (4)(C)) and
sections 302(b) and (5)(C) of the Federal National Mortgage
Association Charter Act (12 U.S.C. 1717(b)(2) and (4)(C)).
---------------------------------------------------------------------------
The Enterprises currently use several different types of credit
enhancements: (1) Private mortgage insurance on individual loans, which
usually covers a percentage of the gross loss, or ``claim amount,''
\152\ (2) seller recourse agreements, which require the seller/servicer
to repurchase loans in the event of default, either for all loan
defaults (unlimited recourse) or for all defaults up to a specified
amount (limited recourse); (3) indemnification, which requires the
seller/servicer to reimburse the Enterprises for losses (either
unlimited or limited) on defaulted loans after final resolution by the
Enterprise; (4) pool insurance, which covers losses on a pool of loans
up to a specified percentage of the aggregate unpaid principal balance
(UPB), usually after private mortgage insurance has been applied; (5)
spread accounts maintained by the Enterprise or a custodian to offset
losses, funded by part of the spread between the interest rate on the
loans in a pool and the coupon passed through to the investor; (6)
collateral pledge agreements under which the Enterprise obtains a
perfected interest in securities held in an account (usually Treasury
securities or mortgage-backed securities), to offset losses on a pool
of loans when a seller/servicer hits certain financial triggers or when
the loans are high risk; and (7) cash accounts funded by the seller/
servicer that are available to offset losses.
---------------------------------------------------------------------------
\152\ The claim amount includes the defaulted principal balance,
unpaid interest, and associated expenses. It does not reflect
subsequent proceeds from the sale of REO.
---------------------------------------------------------------------------
2. Modeling Approach
The stress test calculates the loss coverage provided by credit
enhancements in one of two ways, depending on the credit enhancement
type. Private mortgage insurance, unlimited recourse, unlimited
indemnification, and risk-sharing agreements provide coverage for a
percentage of the loss incurred. The dollar value of these credit
enhancements is not known at the beginning of the stress period because
it depends on the size of the loss that occurs in the future. What is
known is the percentage of the loss that will be covered. Therefore,
these credit enhancement types are referred to herein as ``percent-
denominated'' enhancements. The other credit enhancement types are
referred to as ``dollar-denominated'' enhancements, because the total
coverage provided can be expressed in dollar amounts without knowing
the size of the losses in advance.
The stress test applies the loss coverage provided by credit
enhancements to the loan groups into which individual loans have been
aggregated for modeling efficiency. (See section II. A., Summary of the
Stress Test, for a description of the characteristics that are the
basis for aggregation.) The loss coverage is a weighted average of the
credit enhancements applicable to any loans in the group. In situations
where a loan group is covered by both percent-denominated enhancements
and dollar-denominated enhancements, the two different types of credit
enhancements are applied sequentially. First, the loss severity of a
loan group is reduced by an amount that is determined by the percentage
coverage of the applicable percent-denominated credit enhancements.
Then, the dollar coverage available from dollar-denominated credit
enhancements is applied to the remaining losses on the loan group until
all of the available dollar coverage for that loan group is used up.
This approach permits percent-denominated credit enhancements (such as
private mortgage insurance) to be applied before dollar-denominated
credit enhancements (such as pool insurance) are applied, capturing the
benefits of multi-layered credit enhancements.
[[Page 18151]]
Some dollar-denominated enhancements provide coverage in a dollar
amount that is fixed and known at the time the agreement is executed.
These include pool insurance, limited recourse, limited
indemnification, and cash accounts. Other dollar-denominated
enhancements provide coverage in a dollar amount that is subject to
variation during the term of the agreement. These include spread
accounts and collateral pledge agreements. Changes in these balances
due to reasons other than loss coverage are not modeled. Rather,
balances are treated as cash \153\ and drawn upon after dollar losses
are determined, until the total amount is exhausted.
---------------------------------------------------------------------------
\153\ Although dollar balances for these types may in reality
vary during the stress period, the stress test uses the balance
stated at the beginning of the stress period.
---------------------------------------------------------------------------
Some credit enhancements, namely private mortgage insurance,
recourse, pool insurance, and indemnification, are subject to the
institutional credit risk of the provider, i.e. the risk that the
counterparty providing the credit enhancement will default on its
obligation. Where institutional credit risk is present, the stress test
applies a discount factor, or ``haircut,'' based on the credit rating
of the counterparty.
The haircuts that have been adopted by OFHEO are set forth by
rating category in Table 27:
[GRAPHIC] [TIFF OMITTED] TP13AP99.212
The haircuts reflect the probability that some counterparties will
be unable to meet their obligations during the stress period. Haircuts
become progressively larger as the counterparty rating decreases, with
parties rated BBB or lower and unrated parties receiving the most
severe haircut. The haircut for each rating category is cumulative
rather than additive. It increases for each month of the stress period,
beginning in the first month of the stress test and increasing by equal
amounts (i.e., linearly), until the full amount of the discount is
reached in the 120th month. Table 27 reflects the size of the haircut
at the end of each 12-month period during the stress period. Rating
downgrades are not modeled. Instead, deterioration in the financial
condition of counterparties due to the stressful environment is
reflected in the linear increase of the haircuts.
3. Comments and Alternatives Considered
In the ANPR, OFHEO requested comments on how to calculate the loss
coverage provided by credit enhancements and on what assumptions to
make about the scope of coverage and the failure of counterparties
during the stress period. These and other issues, relevant comments
received, and OFHEO's rationales for the selected approaches are
discussed below.
a. Modeling Approach
ANPR commenters suggested a variety of modeling approaches. MICA
stated that the capital requirements for the Enterprises should be
consistent with capital requirements for banks and thrifts and reflect
the underlying product risk associated with each class of mortgage-
related assets. MICA recommended that OFHEO assign relative ``capital
relief'' values to ``the three allowable credit enhancements'' \154\
based on the quantity and quality of the credit enhancement. MICA
further recommended that OFHEO consider mortgage insurance provided by
a company with at least a AA claims-paying rating and providing at
least the minimum coverage required by the Enterprises' charters as the
``benchmark credit enhancement.'' The benchmark credit enhancement
should receive the ``maximum amount of capital relief,'' and other
forms of credit enhancement should receive values relative to this
benchmark, based on the quality and quantity (i.e. the amount of the
loss it covers) of the enhancement. (See section III.C.3.c.,
Discounting for Counterparty Risk for a discussion of MICA's comments
related to the quality of the credit enhancement.) MICA views this
approach as consistent with risk-based requirements for banks and
thrifts, which require uninsured high-LTV
[[Page 18152]]
loans held in portfolio to have twice as much capital as high-LTV loans
that are privately insured.
---------------------------------------------------------------------------
\154\ OFHEO interprets ``three allowable credit enhancements''
as a reference to the three types of credit enhancement mentioned in
the Charter Act exception to the prohibition on purchasing loans
with LTVs in excess of 80 percent.
---------------------------------------------------------------------------
Freddie Mac suggested a two-step process similar to the process it
uses in its internal models for pricing transactions. Freddie Mac first
estimates the value of the credit enhancement by estimating the
proportion of default losses that would be covered, and then discounts
the estimated value to reflect the institutional credit risk of the
provider, if any. Although Freddie Mac`s credit enhancement valuation
process occurs at the transaction level for pools of mortgages, Freddie
Mac suggested that such a transaction-level approach might not be well
suited for OFHEO's stress test. Rather, it recommended aggregating
credit enhancements into categories before applying the two-step
process. Freddie Mac further recommended that private mortgage
insurance be modeled in connection with the modeling of loss
severities. Other types of credit enhancements, Freddie Mac suggested,
could be converted to ``collateral-equivalent'' amounts and, after
discounting for applicable institutional credit risk, aggregated into a
large collateral-equivalent pool and used to offset stress test losses
dollar for dollar. Freddie Mac made specific recommendations for
collateral-equivalent conversions: collateral pledge agreements and
spread accounts should be included on a dollar-for-dollar basis and
future inflows to spread accounts should be estimated based on the
weighted average life (WAL) of the pool; \155\ pool insurance should be
included to the policy limit, i.e. the percentage limitation multiplied
by the original UPB; and recourse and indemnification agreements should
be treated as if 100 percent of the losses from mortgage defaults in
the applicable pools were covered until such time as the seller/
servicer failed.
---------------------------------------------------------------------------
\155\ This could be done by multiplying the WAL by the average
yearly spread going into the spread account and then by the UPB.
---------------------------------------------------------------------------
The approach adopted by OFHEO is similar in many respects to the
approach suggested by Freddie Mac. Like Freddie Mac's approach, it
estimates the probable coverage of credit enhancements and discounts
for counterparty risk where it is present. The value of private
mortgage insurance and other forms of credit enhancements that cover a
percentage of loss is estimated in connection with loss severities, as
suggested by Freddie Mac. The approach adopted by OFHEO differs from
the approach suggested by Freddie Mac in some of the details of how
credit enhancement coverage is estimated and how discounts for
counterparty risk are calculated. These differences are discussed
further below.
b. Aggregation
A threshold issue for OFHEO was whether to track and model each
credit enhancement with the loan or pool to which it relates or to use
some level of aggregation for credit enhancements to increase modeling
efficiency. Tracking and modeling each individual credit enhancement
agreement with the particular loan or pool to which it is related would
yield the most precise estimate of the value and behavior of credit
enhancements, but would make the model very complex. Aggregating credit
enhancements for efficiency in modeling, on the other hand, gives rise
to ``cross support,'' which overestimates the amount of credit
enhancements that would actually be used to offset losses. ``Cross
support'' means that credit enhancements provided on a particular loan
or pool are available to offset losses on another loan or pool, when in
practice they would be available only to offset losses on the
particular loan or pool for which they were provided and would be
partially unused if losses were lower than the amount of the coverage.
However, in a model that aggregates credit enhancements and applies
them to loan groups, the unused portion of a credit enhancement is
available to cover losses in the same loan group. The greater the
aggregation of credit enhancements in the stress test, the more cross
support occurs, and the more the estimated value of the credit
enhancements is overstated. Aggregation up to a very high level can
introduce an unacceptable level of cross support.
OFHEO considered converting each credit enhancement type to a
dollar-equivalent amount, aggregating these amounts across all credit
enhancement types into a single pool of collateral-equivalent dollars,
and applying them dollar for dollar against stress test losses. While
this approach is simpler and would have required less intensive
tracking, it would permit an unacceptable level of cross-support by
credit enhancements of different types and for different loan groups.
Just as importantly, this approach would not have produced accurate
results for the coverage associated with percent-denominated credit
enhancements, such as private mortgage insurance. The dollar amount of
coverage of these credit enhancements cannot be calculated until losses
are determined. These losses can only be calculated during the course
of the stress period; they are not known at the beginning of the stress
period.
The approach adopted by OFHEO strikes a balance between the
benefits of simplicity and efficiency and the benefits of precision
while imposing minimal regulatory burden. By estimating the coverage
provided by each type of credit enhancement on the basis of loan
groups, tracking credit enhancements for each loan group can be
accomplished efficiently. The large number of loan groups used by the
stress test minimizes cross support between different types of credit
enhancements, loans, and time periods.
c. Discounting for Counterparty Risk
Another issue faced by OFHEO was whether and how to take into
account the risk that the counterparty's ability to perform on the
credit enhancement agreement would be affected by the conditions of the
stress test.
OFHEO received a number of suggestions on the treatment of
counterparty risk in response to the ANPR. Freddie Mac, MICA, and ACB
recommended incorporating an assumption that some of the counterparties
would fail during the stress period and suggested that OFHEO look to
private rating agencies for guidance. ACB suggested that the OFHEO
analysis of the actual coverage provided by mortgage insurance during
the stress period could be ``piggybacked'' on S&P's analysis. ACB
further stated that OFHEO could make reasonable adjustments to align
the worst-case scenario in S&P's stress test with that in the OFHEO
analysis, and that it would not be necessary to extend the analysis
beyond private mortgage insurers.
As noted earlier, MICA recommended a matrix for determining
``capital relief'' for credit enhancements relative to a benchmark
credit enhancement. One dimension of the recommended matrix is the
credit rating of the counterparty, reflecting an assumption that the
values assigned to various credit enhancements should reflect a
differentiation on the basis of the provider's claims-paying rating.
However, MICA's recommendation that OFHEO give ``maximum capital
relief'' (at least 50 percent of the normal capital charge) to a AA-
rated insurer providing at least the minimum coverage required by the
Enterprises' charters appears to be equivalent to a recommendation that
AA-rated counterparties not be discounted at all.\156\ MICA asserted
that
[[Page 18153]]
this recommendation is supported by the historical default experience
for corporate bonds in the 1970-89 period, particularly the 0.9 percent
default rate for AA-rated bonds.\157\ From this MICA concluded that
99.1 percent of mortgage insurance would be available to the
Enterprises during the stress period.
---------------------------------------------------------------------------
\156\ The risk-based capital requirements for banks and thrifts
are not determined by a statutorily prescribed stress test but by
establishing a standard capital charge for all assets that is
expressed as a fixed percentage of the face amount of the asset.
Capital relief for particular assets is achieved by risk weighting
them at less than 100 percent of the face amount. Risk-based capital
regulations for banks and thrifts risk-weight mortgage loans at 50
percent of the UPB. In a stress test regulation, the most favorable
capital treatment is achieved by giving full credit for the credit
enhancement without any discount.
\157\ ``Approach to Rating Residential Mortgage Securities,''
Moody's Investor Service, April 1990.
---------------------------------------------------------------------------
Freddie Mac recommended that evaluation of counterparty risk be
based on the probable length of time an institution would continue
meeting its loss-paying obligations in the stress period, which would
be determined by the institution's rating at the beginning of the
stress period. This method, Freddie Mac asserted, is similar to one
used by Moody's. Specifically, AAA-rated companies would be assumed to
cover all obligations for the entire ten-year stress period. AA-rated
companies would be assumed to cover all obligations for seven years and
none thereafter, A-rated companies for five years, and companies rated
BBB and lower, only three years. Freddie Mac also recommended that
institutions that are required to post collateral under a collateral
pledge agreement be ranked with AAA-rated institutions. For recourse
and indemnification agreements, Freddie Mac suggested that OFHEO could
assume the agreement would last until the institution failed, a time
determined by the institution's rating. It noted, however, that a
similar effect could be achieved by adjusting the loss severities based
on institution ratings, where the adjustment to loss severity would be
lower for a higher institutional rating. However, Freddie Mac cautioned
that if this approach were used, the difference between the present-
value cost of losses occurring at the end of the stress period and
losses occurring at the beginning of the stress period would have to be
taken into account. That is, an institution that honors its recourse
agreement for the first five years of the ten-year stress period would
pay out much more than half of the present value of the losses.
Only one commenter suggested that credit enhancements having
counterparty credit risk not be discounted for the risk. The MBA
expressed concern about the burden it would place on the Enterprises to
determine the financial strength of third parties and suggested that
credit enhancements need not and should not be discounted for credit
risk of the counterparty. The reasons cited were three. First, the
Enterprises generally accept credit enhancements only from well-
capitalized companies. Moreover, the Enterprises are in a good position
to evaluate the counterparty's financial strength,\158\ and the seller/
servicer agreement often provides added protection from default on
repurchase or indemnification obligations. Second, an assessment of
counterparty credit risk is reflected in guarantee fees, which can be
adjusted with each commitment. And third, mortgage insurers are
nationally rated by recognized organizations that routinely adjust
ratings based on changes in financial status. As a result, trends in
their financial health can be monitored easily. The MBA urged OFHEO to
ground its assumptions and conclusions in historical experience and
``real world'' conditions, which, in its view, argue for not
discounting credit enhancements for counterparty risk.
---------------------------------------------------------------------------
\158\ This results, MBA noted, from close relationships between
the Enterprises and seller/servicers based on frequent marketing
contacts, Enterprise auditing activities, and lender reporting
obligations.
---------------------------------------------------------------------------
OFHEO believes that some counterparty failure would be likely under
the stressful conditions imposed by the stress test and that
discounting for counterparty credit risk is necessary to avoid
overstating the effect of credit enhancements in covering losses. The
statutorily required benchmark stress period is considerably more
severe than the national historical experience of corporate bonds cited
by MICA. Also, as noted by Anthony Yezer, Professor of Economics at
George Washington University, the failure of private mortgage insurers
was important in the collapse of the thrifts in the 1930s.
Although the stress test reflects assumptions about the claims-
paying abilities of counterparties during the stress period that are
similar to Freddie Mac's, OFHEO did not adopt Freddie Mac's assumption
that counterparties would pay 100 percent of their obligations as long
as they paid at all. In OFHEO's judgment, this assumption is
inconsistent with the pattern of counterparty defaults on obligations
that one would expect during a stressful period and inconsistent with
the pattern of defaults observed in the past. For example, Moody's
study of corporate bond defaults \159\ showed that cumulative defaults
in each of the various ratings categories increased gradually over
time. Also, it is likely that the primary market and credit enhancement
counterparties would be affected by the stress test conditions
relatively early in the stress period. Freddie Mac's approach would not
capture this early impact. If mortgage losses were to occur during the
first half of the stress period, the importance of reductions in credit
enhancements due to counterparty risk would be understated because, as
noted by Freddie Mac, mortgage losses occurring during the first half
of the stress period constitute much more than half of the present
value of total losses. Therefore, credit enhancements offsetting those
losses would be more valuable. A more realistic assumption is that the
rate of counterparty defaults would increase gradually during the
stress period.
---------------------------------------------------------------------------
\159\ ``Historical Default Rates of Corporate Bond Issuers,
1920-1997,'' Moody's Investors Service, February 1998.
---------------------------------------------------------------------------
OFHEO did not adopt Freddie Mac's recommendation to treat seller/
servicers who are required to post collateral when certain financial
triggers are met \160\ the same as AAA-rated institutions. Freddie Mac
contends that the existence of these agreements would provide coverage
equivalent to a AAA-rated credit enhancement. However, whether
collateral would actually be posted when required is an additional
source of counterparty risk and whether that collateral would provide
coverage equivalent to a AAA-rated credit enhancement is difficult to
evaluate in a regulatory context. Such an evaluation would require
OFHEO either to develop the capacity to rate each seller/servicer with
a collateral pledge agreement and the impact of the agreement on the
seller/servicer's rating, or to require the Enterprises to obtain
public ratings for such seller/servicers that take these agreements
into account. In light of the small impact that this degree of
precision is likely to have on the capital requirement, OFHEO believes
that developing such a rating capacity is not an appropriate use of
regulatory resources, and that requiring the Enterprises to obtain
public ratings would impose an undue regulatory burden. Consequently,
the proposed stress test does not model the value of collateral pledge
agreements. Instead, it only models coverage provided by collateral
that is already available in an Enterprise or third-party account.
---------------------------------------------------------------------------
\160\ Seller/servicer agreements may include such a requirement
when there is a decline in the institution's rating or a decline in
its capital levels below a specified amount.
---------------------------------------------------------------------------
This treatment is consistent with the treatment of such agreements
under OFHEO's minimum capital regulation. Collateral is not recognized
for purposes
[[Page 18154]]
of satisfying the minimum capital standard unless it is actually held
and legally available to absorb losses. Also, to be consistent with the
minimum capital restrictions on the forms of collateral that are
acceptable, the proposed stress test will give credit for the coverage
provided by collateral only if it is among the following types: cash on
deposit; securities issued or guaranteed by the central governments of
the OECD-based group of countries,\161\ United States Government
agencies, or United States Government-sponsored agencies, and
securities issued by multilateral lending institutions or regional
developments banks.
---------------------------------------------------------------------------
\161\ The OECD-based group of countries comprises all full
members of the Organization for Economic Cooperation and Development
and countries that have concluded special lending arrangements with
the International Monetary Fund (IMF) associated with the IMF's
General Arrangements to Borrow, but excludes any country that has
rescheduled its external sovereign debt within the previous five
years.
---------------------------------------------------------------------------
In determining the size and timing of the discounts (haircuts) to
the value of the credit enhancements, OFHEO considered Moody's study of
corporate bond default rates and methodologies used by S&P and Duff &
Phelps (D&P). Moody's analysis of corporate bond issuers from 1920 to
1997 \162\ showed cumulative default rates over various time horizons
for each rating category. The average ten-year cumulative default rate
over the entire period was 1.17 percent for Aaa issuers, 3.32 percent
for Aa issuers, 3.87 percent for A issuers, 8.08 percent for Baa
issuers. These data suggest that the ten-year cumulative default rate
roughly doubles for each one-level drop in rating category. Defaults
for Aa issuers were higher relative to those for Aaa and A issuers than
this doubling relationship would suggest. However, Aa issuers from the
mid-1970s forward had ten-year cumulative default rates that were much
lower relative to issuers in other rating categories.
---------------------------------------------------------------------------
\162\ ``Historical Default Rates of Corporate Bond Issuers,
1920-1997,'' Moody's Investors Service, February 1998.
---------------------------------------------------------------------------
The Moody's approach and the approach recommended by Freddie Mac is
a survival approach in which it is assumed that an institution meets
100 percent of its obligations for as long as it survives, and relative
risk is expressed as the number of years an institution survives. The
approach used by S&P and D&P \163\ is a haircut approach in which it is
assumed that institutions will meet some, but not all, of their
obligations, and the haircut is the percent of obligations they will
fail to meet. Specifically, S&P discounts the claims-paying ability of
mortgage insurers in a AA stress level environment by 20 percent for
AA-minus-rated mortgage insurers, 50 percent for A-rated mortgage
insurers, and 60 percent for A-minus-rated mortgage insurers. D&P
discounts mortgage insurers in a AAA stress level environment by 35
percent for AA-rated reinsurers, 70 percent for A-rated reinsurers, and
100 percent for BBB-rated reinsurers. For S&P, the haircuts apply in
full from the second year of the stress period. Also, the haircut is
related to the stress level of the environment, and an insurer with a
rating equal to or greater than the stress level is not discounted.
---------------------------------------------------------------------------
\163\ ``S&P's Structured Finance Criteria,'' Standard & Poor's
Corporation, 1988; ``Evaluation of Mortgage Insurance Companies,''
Duff & Phelps, November, 1994.
---------------------------------------------------------------------------
Moody's corporate bond study shows that the cumulative default
curves for companies with ratings of BBB and above were essentially
linear.
[GRAPHIC] [TIFF OMITTED] TP13AP99.370
OFHEO's approach to applying haircuts is similar to S&P's and
D&P's, but differs in three ways. First, the stress test does not apply
the full amount of the haircut immediately but applies a haircut that
increases each month until reaching the full amount in the 120th month.
This reflects the general industry view that defaults increase
gradually in a stress scenario. Further, as illustrated by the graph in
Figure 2, the linear growth specification of the stress test is a
reasonable one in light of actual historical patterns of default.
Second, the stress test haircuts are in no case as low as zero and in
no case as high as 100 percent. This reflects historical default
patterns, which suggest that counterparties or issuers in each rating
category would pay at least some claims, and no rating category would
be immune from any claims-paying defaults. With respect to the absence
of a rating category with zero defaults, Moody's data show that, in a
difficult but far from severe environment, 3.2 percent of issuers
[[Page 18155]]
rated Aaa at the beginning of 1983 defaulted within 10 years. Third,
the stress test haircuts are not tied to the stress level. While
OFHEO's NPR 1 showed credit stress at roughly a AA+ level, the stress
test as a whole does not translate to any particular level because
OFHEO's methodology as required by the 1992 Act differs in several key
respects from that used by rating agencies.
Although OFHEO considered developing a probabilistic survival
function for counterparties that would provide an estimate of failure
in each year of the stress period, such a methodology would be
difficult to specify, implement, and replicate, especially if recovery
rates on bankrupt counterparties were modeled. OFHEO concluded that,
short of a probabilistic function, imposing a linearly increasing
haircut on all counterparty credit enhancement proceeds through the
entire stress period would be the most representative of all the other
options of how the rate of counterparty defaults would increase during
the ten-year stress period.
The size of the haircuts proposed for the stress test, ten percent
for AAA-rated companies, 20 percent for AA-rated companies, 40 percent
for A-rated companies, and 80 percent for BBB-rated companies, are far
more severe than recent default experience but less severe than
Depression-era experience. They are about six to ten times the severity
of average ten-year cumulative defaults during 1920-1997 in the Moody's
analysis. The haircuts double for each drop in rating category,
consistent with the Moody's bond default analysis. Some default occurs
among AAA-rated companies, while BBB-rated company defaults are not 100
percent.
OFHEO's approach is transparent, easily replicated, and consistent
with industry practice. It draws on the best aspects of S&P's approach
to modeling mortgage insurer performance, and Moody's corporate bond
study in applying company defaults over time. It also recognizes that,
while the impact of the stress test environment on Enterprise losses
might not be large in the first two years of the stress period, the
primary mortgage market (i.e., the seller/servicer counterparties)
likely would feel the impact of a stressful environment almost
immediately.
d. Unrated Seller/Servicers
OFHEO considered whether unrated seller/servicers should be treated
the same as other unrated counterparties or whether they should be
treated differently because of their close relationships with the
Enterprises.
Both Freddie Mac and MBA argued that even though seller/servicers
are typically unrated, the close relationship between the Enterprise
and its seller/servicers enables the Enterprise to monitor their
financial strength. Freddie Mac stated that the seller/servicer
agreement provides added protection against default on recourse and
indemnification obligations because it gives Freddie Mac the right to
the servicing of all Freddie Mac loans then serviced by the institution
in the event of default on these obligations. Freddie Mac asserted that
the value of the servicing is likely to cover a substantial portion of
the defaults covered by a seller/servicer recourse agreement.\164\ For
these reasons, Freddie Mac considers all sellers/servicers to be at
least BBB for purposes of evaluating institutional credit risk and
urged OFHEO to consider the added layers of protection provided by the
servicing rights.
---------------------------------------------------------------------------
\164\ Freddie Mac estimates that these servicing rights are
normally worth about 25 basis points of income per year, and can be
sold to another servicer for 100 to 150 basis points.
---------------------------------------------------------------------------
The stress test treats unrated seller/servicers, like other unrated
counterparties, the same as it treats BBB counterparties, which is
consistent with the thrust of Freddie Mac's ANPR comments. Although
OFHEO does not explicitly price the added layer of protection provided
by mortgage servicing rights in its stress test, this added layer of
protection was considered as a factor in deciding that unrated
counterparties should be treated as BBB. OFHEO believes that any
imprecision resulting from assigning unrated seller/servicers to the
BBB or lower rating group would have a small impact on the resulting
capital requirement. Seller/servicer recourse represents a small
percentage of the credit enhancements used by the Enterprises. In
addition, the Enterprises' largest customers tend to have public
ratings.
Although the Enterprises assign internal ratings to seller/
servicers, OFHEO did not use these ratings for three reasons. First,
these ratings and the methodology for developing the rating are
proprietary information and not publicly available. Therefore, they
cannot be included in the regulation or used by third parties to
evaluate the risk-based capital requirement. Second, each of the
Enterprises has developed its own unique rating system. These rating
systems may result in different ratings of the same parties. One of the
underlying requirements of this regulation is the development of a
capital requirement that is applied uniformly to both Enterprises. This
requirement cannot be met if different rating systems are applied to
each Enterprise. Finally, using such ratings without independent
validation by OFHEO would compromise the independence of the regulatory
process.
e. Fluctuations in Value
The dollar value of some credit enhancements, such as spread
accounts and securities deposited in an account under collateral pledge
agreements, fluctuate over time, for reasons other than withdrawals to
cover losses. Spread accounts are funded by a portion of each loan
payment and hence increase in value as loan payments are made.
Securities deposited in an account under collateral pledge
agreements,\165\ which are marked to market periodically, fluctuate in
value due to movements in interest rates during periods that fall in
between the marks to market. In addition, posting requirements of
collateral pledge agreements can cause additional collateral to be
deposited to the account.
---------------------------------------------------------------------------
\165\ As stated earlier, the stress test recognized the coverage
provided by collateral pledge agreements only if collateral has
actually been posted and resides in an account as of the beginning
of the stress period. Otherwise, collateral pledge agreements are
not modeled in the stress test.
---------------------------------------------------------------------------
The stress test does not model these fluctuations. Rather, it uses
the dollar value of spread accounts, cash accounts, and collateral
posted under collateral pledge agreements on the first day of the
stress period and draws on this dollar amount throughout the stress
period to cover losses. Modeling fluctuations in the value of
collateral posted under collateral pledge agreements would have added a
level of complexity that is not justified by the incremental precision
that would be gained. Similarly, the stress test does not model the
accumulation of interest in the spread account according to the terms
of the spread account agreement because this would have introduced a
level of complexity that is not justified by the probable impact on the
ultimate capital requirement.
Freddie Mac suggested that OFHEO estimate future inflows by
multiplying the weighted average life (WAL) of the mortgage pools by
the average yearly spread going into the spread account and then by the
UPB. However, such an approach would also have made the stress test
excessively complex. Loans covered by a spread account agreement may be
in different loan groups in the stress test, and determining the WAL of
[[Page 18156]]
all the loans covered by each spread account would require tracking
each spread account loan and processing spread account characteristics
at the transaction level.
OFHEO will continue to monitor the relative volume of spread
accounts and collateral pledge agreements and consider whether an
amendment to the regulation is needed if it should appear that the
impact on the capital requirement might be significant.
f. Credit Enhancement on High LTV Loans
Certain credit enhancement types used by the Enterprises are not
mentioned in the Charter Acts' exceptions to the prohibition on
purchasing single family loans with LTVs in excess of 80 percent,
namely spread accounts, collateral pledge agreements, cash accounts,
pool insurance, and indemnification. This fact raised the issue of
whether the stress test should take them into account when they are
intended to satisfy the statutory requirement for credit enhancement on
loans with LTVs in excess of 80 percent. In its comment letter, Freddie
Mac argued that an expansion of the list of recognized credit
enhancements to include collateral pledge agreements, spread accounts,
and indemnification would be consistent with the intent of Congress in
giving the OFHEO Director discretion to make reasonable assumptions
about factors that would affect the severities of loss on mortgage
defaults, including ``the value of mortgage insurance [and] the value
of various forms of credit enhancements such as recourse agreements,
collateral, and spread accounts.'' \166\ MICA, on the other hand,
argued that only the three types mentioned in the statutory exceptions
should be considered.
---------------------------------------------------------------------------
\166\ H.R. Rep. No.102-206, at 67 (1991).
---------------------------------------------------------------------------
Although OFHEO recognizes that some types of credit enhancements
not expressly referenced in the Charter Acts may provide equal or
superior loss protection, OFHEO does not believe that they satisfy the
statutory requirement for credit enhancements for single family loans
with LTVs in excess of 80 percent. OFHEO does not concur with Freddie
Mac that the legislative history of the 1992 Act gives OFHEO the
latitude to expand the list of statutorily authorized credit
enhancements for single family loans with LTVs in excess of 80 percent.
OFHEO believes that taking into account credit enhancements not
expressly referenced in the Charter Acts when they are used to satisfy
the statutory credit enhancement requirement for single family loans
with LTVs in excess of 80 percent would undermine OFHEO's efforts to
ensure that the Enterprises operate within the Charter Acts.
g. Scope of Coverage
The ANPR asked for comments on how the regulation should address
the scope of coverage provided by credit enhancements. Freddie Mac, the
only commenter on this question, stated that all credit enhancements
except private mortgage insurance can be assumed to cover all loss
elements, including loss of property value, lost interest, real estate
commissions, attorney fees, taxes, and preservation costs, where as
private mortgage insurance sometimes excludes certain expenses after
the property becomes REO.
Based on an analysis of available information, OFHEO proposes to
make credit enhancements coverage available for all types of losses
associated with stress test defaults. The benchmark data reveal that
loss severities before credit enhancements were applied for single
family loans in the benchmark time and place were consistently in the
50 percent to 60 percent range. At the same time, private mortgage
insurance coverage typically ranged from 12 percent to 30 percent
coverage of the gross claim amount. Since the severities far exceed the
coverage of private mortgage insurance, the stress test assumes that
the private mortgage insurance would be used up covering expenses that
the mortgage insurance typically covers, and that the REO-related
expenses would be reflected in the uncovered losses.
h. Termination of Private Mortgage Insurance
Modeling private mortgage insurance required a determination of how
to treat the potential for termination of mortgage insurance while the
loan is outstanding. Termination occurs either because the borrower
exercises an option to cancel the insurance when the equity in the loan
reaches a predetermined threshold, or because cancellation is automatic
under the provisions of the recently enacted Homeowners Protection Act
of 1998.\167\ For loans originated before the July 1999 effective date
of the Homeowners Protection Act, termination resulting from the
borrower's exercise of the right to cancel the insurance when
sufficient equity in the loan is attained presents a difficult issue,
because data on this phenomenon are scarce, and there is an
insufficient basis on which to draw firm conclusions. OFHEO considered
three options: (1) assume that borrowers do not exercise this right
when they are eligible; (2) assume all borrowers exercise this option
when they become eligible; or (3) assume some percentage of borrowers,
less than 100 percent, exercise this option when they become eligible.
---------------------------------------------------------------------------
\167\ Pub. L. No.105-216, 112 Stat. 897-910 (12 U.S.C. 4901-
4910).
---------------------------------------------------------------------------
After considering these options, OFHEO concluded that the first
option was the preferred option because it is the option likely to
produce the least distortion. The second option would understate the
amount of credit enhancement available and the third would require an
assumption based on very sparse data. Although assuming that insurance
is not terminated may be a source of some imprecision, the impact of
such imprecision is not likely to be significant in determining capital
needed under the stress test. The loans most likely to default are
those loans with high current LTV ratios, which will not be eligible
for termination of private mortgage insurance because of the high LTVs.
Conversely, those loans with low enough current LTV ratios to be
eligible for termination are much less likely to need the coverage, and
whether it is unused or is assumed to be terminated will make little
difference. The largest potential for error is with loans with high
original LTV ratios that have aged prior to the stress test just to a
point where coverage can be terminated. OFHEO will monitor this issue
and consider proposing an amendment to the regulation if another option
appears to be more appropriate.
The Homeowners Protection Act provides that mortgage insurance will
terminate automatically when the loan balance is scheduled to reach 78
percent of the original value of the property securing the loan,
provided payments on the loans are current. For loans that do not meet
the LTV test and for high-risk loans with original principal balances
that do not exceed the conforming loan limit, mortgage insurance will
terminate when the loans reach the mid-point of their amortization
periods if payments are current. The Enterprises will publish
guidelines to describe high-risk loans. OFHEO proposes to apply the
provisions of the Act by eliminating mortgage insurance coverage in
calculating loss severities for loans that reach 78 percent of their
original value during the stress period or at the midpoint of their
amortization periods for ``high risk'' loans, as defined by the
Enterprises.
[[Page 18157]]
D. Liabilities and Derivatives
The Enterprises issue a variety of debt instruments that comprise
their liability portfolios. To understand the types of liabilities
issued by the Enterprises it is useful to group the liabilities into
categories based on similar characteristics related to the instrument's
coupon type, optionality, or other structuring features. The
liabilities issued by the Enterprises are primarily one of three coupon
types: fixed-rate, floating-rate, or zero-coupon. The Enterprises use
these different types of coupons to manage both their exposure to
interest rate risk and their cost of funding. The optionality of a
financial instrument refers to whether that instrument contains an
embedded option--in the case of the Enterprises liabilities, generally
a call option. The embedded call option gives the Enterprises the
opportunity to pay off (call) the debt, at a time prior to its
contractual maturity. The Enterprises issue a mix of callable and non-
callable (bullet) debt in order to manage their exposure to the
prepayment risk inherent in their retained mortgage and mortgage
security portfolios.
The Enterprises also issue liabilities that have unique structuring
features, such as complex principal, coupon, or optionality
characteristics. An example of a complex liability is a Euro discount
note. To the extent that these notes are issued in foreign currencies,
the Enterprises are exposed to foreign exchange risk, which is offset
with hedging transactions at the time the discount notes are issued. An
example of a liability with complex coupon characteristics is an
inverse floater. For example, this instrument may pay a fixed rate of
interest for a given period of time and then revert to an interest
payment based on the formula 12 percent less six month LIBOR. In this
case, the Enterprises incur higher interest costs as LIBOR decreases.
In most situations, the complex risk characteristics of these
liabilities are hedged at the time of issuance, leaving the Enterprise
with synthetic ``plain vanilla'' liabilities, which have the coupon and
option features of a more typical Enterprise liability. These
liabilities generally are used by the Enterprises to obtain funds at a
lower net cost than could be obtained by issuing simpler forms of debt.
In addition to the types of liabilities discussed above, the
Enterprises also provide investment vehicles, termed Guaranteed
Investment Contracts (GICs), to various institutions that have specific
cash flow requirements or need flexibility in making cash withdrawals.
They comprise a very small percentage of the Enterprises' liabilities.
GICs can pay or accrue interest. Their principal balances can increase,
decrease or remain the same.
The Enterprises, like most large financial institutions, use
derivatives to help manage the interest rate risk of their assets and
liabilities. The term ``derivatives'' covers a broad range of
instruments, the value of which is based on or linked to (i.e.,
``derived'' from) another instrument or a financial market such as
stocks, interest rates or currencies. A common derivative is an
interest rate swap, which derives its value from the changes in value
of interest rates paid on various types of debt instruments.
Derivatives can be used to hedge the unusual or complex risk
characteristics of individual debt instruments, such as the complex
structured liabilities described above. They also can be used to
rebalance the interest rate risk of an entire portfolio. In short,
derivatives, like most financial instruments, can either add or reduce
various types of risk. The risk-based capital regulation, therefore,
must account for derivatives in order to reflect accurately the risk
profile of the Enterprises.
In developing an approach for modeling the cash flows of the
Enterprises' liabilities and derivatives, OFHEO had to address four
issues discussed below: (1) should liabilities and derivatives be
modeled at the instrument level or should they be aggregated in some
manner; (2) how should instruments linked to foreign currencies or
unusual risk factors be modeled; (3) how should callable debt and
cancellable derivatives be modeled; and (4) how should the stress test
account for the risk of derivative counterparty defaults?
1. Modeling Methodology
The first issue for OFHEO was whether to model liability and
derivative cash flows at the instrument level or to aggregate
individual instruments with similar terms and risk characteristics and
model the aggregated cash flows based upon average maturities, coupons,
options, and other features. In response to an ANPR question about how
OFHEO should simulate gains and losses on derivative activities,
Freddie Mac suggested that the underlying instruments should be
modeled. Likewise, Freddie Mac's discussion of liabilities in its
comments assumes that most liability instruments will be modeled
individually. The only other comment was ACB's suggestion regarding
accounting for the risk of counterparty default. ACB's recommendation
that the stress test ``haircut'' (meaning reduce by a percentage)
derivative positions when they were ``in the money'' (meaning the
derivatives have a net positive value to the Enterprises) would require
modeling cash flows of derivatives individually.
The issue of modeling liabilities and derivatives on an aggregated
versus instrument level usually requires a trade-off between accuracy,
model complexity, and information system resources. In most cases, the
model for generating cash flows uses the same types of information for
an individual instrument as it would for a group of similar
instruments. For this reason, OFHEO's information system resources are
capable of processing the large number of individual liabilities and
derivatives in a reasonable amount of time. Therefore, OFHEO proposes
to model the cash flows of all existing types of liabilities and
derivatives individually, except certain instruments that have terms or
risk characteristics based on a foreign currency, which are discussed
below as a separate issue.
As with most other liabilities, the stress test will model GICs
individually. However, given the variety of their terms and purposes,
it was necessary to simplify the cash flow model for these instruments.
The stress test models each GIC as if it pays out its specified
interest on the starting balance amount over the entire stress period,
unless the GIC includes an explicit maturity date. In the latter case,
the stress test pays interest only until the maturity date, at which
point it pays out the total principal.
2. Foreign Currency Linked or Unusual Instruments
The second liabilities-related issue arises because, from time to
time, the Enterprises issue foreign currency-denominated debt and
structured notes that are linked to a foreign currency. As discussed
above, the Enterprises currently hedge all foreign currency-linked
securities with derivatives or other financial instruments, resulting
in synthetic securities denominated in U.S. dollars. Freddie Mac, the
only ANPR commenter to address this issue, recommends modeling foreign
currency-linked transactions differently from other instruments,
explaining that ``hedge cash flows or the netted cash flows need to be
calculated * * *.''
OFHEO agrees that currency-linked securities and the associated
hedging instruments are different from other types of liabilities and
derivatives of the Enterprises in that the cash flows of the individual
instruments are linked to changes in currency values. OFHEO also
[[Page 18158]]
recognizes that, in current practice, the Enterprises issue a limited
volume of currency-linked instruments and transfer all currency risk to
third parties by hedging instruments. Further, with the exception of
debt linked to foreign currency, the Enterprises have not issued
liability instruments that were linked to indices or values (such as
commodities or stock prices) that are not projected in the stress
test.\168\
---------------------------------------------------------------------------
\168\ However, wherever the terms ``foreign currency'' or
``currency'' are used, they should be read to include any unit or
value, except those interest rate indices that are included in the
stress test, in which debt or derivatives may be denominated or to
which such instruments may be linked.
---------------------------------------------------------------------------
OFHEO concurs with Freddie Mac's comments that where all the
currency risk is hedged, by swapping the foreign currency payments into
dollars, the stress test could calculate the cash flows by creating a
single synthetic liability, denominated in dollars and paying the net
amount due under the related transactions. The stress test, therefore,
applies that approach to instruments that are fully hedged. However, in
the event that OFHEO finds that the foreign currency risk on any
liability or derivative instrument has not been transferred fully to a
third party, the stress test models the cash flow on such instruments
as follows.
The stress test creates significant losses in unhedged currency
positions in both the up-rate and down-rate scenarios. In the up-rate
scenario, the stress test applies an exchange rate that increases the
value of the foreign currency against the dollar by the same percentage
that interest rates increase. For example, if the ten-year CMT shifts
up by 50 percent, then the foreign currency value is shifted up by 50
percent against the dollar for the up-rate scenario.\169\ The effect in
this example would be that the Enterprise would be paying 50 percent
more dollars due to the unhedged exchange rate shift.
---------------------------------------------------------------------------
\169\ Shifting the value of the other currency up 50 percent has
effect of decreasing the value of the dollar against that currency
by \1/3\. In other words, one could buy the same amount of dollars
with only \2/3\ the amount of other currency.
---------------------------------------------------------------------------
A different adjustment is applied in the down-rate scenario. In
that case, the stress test decreases the exchange rate of the dollar
proportionately with the decline in the ten-year CMT, creating a
decrease in the value of the dollar similar to that in the up-rate
scenario. Thus, a downward shift in the ten-year CMT of 50 percent
would be associated with a shift down of 50 percent in the exchange
rate of the dollar. The effect in this example is that the Enterprise
would be paying twice as many dollars due to the unhedged exchange rate
shift.
This approach is simple, conservative and reasonable. The stress
test recognizes that there can be substantial risk associated with
unhedged positions in foreign currencies or other indexes or values to
which instruments can be linked, but that it would be impractical for
OFHEO to develop indexes for foreign currencies and all other values to
which liabilities or derivatives could be linked. The exchange rate in
the up-rate scenario is not based upon a model or an economic
prediction, but does reflect a recognition that there have been
occasions in the past where the dollar has declined in value as CMT
rates have been increasing. Likewise, the dollar has also declined at
times when CMT rates have decreased. Therefore, it is appropriate in a
stress test to assume that the dollar moves in an unfavorable direction
in both scenarios, to avoid creating a windfall to the Enterprises and
to ensure significant financial stress in both scenarios. Moreover,
OFHEO does not anticipate at this time that the Enterprises will be
issuing foreign currency or unusual debt derivatives without using
appropriate and complete hedges. If the Enterprises do alter their
current businesses to enter into such debt, OFHEO will consider at that
time whether a different treatment for the instruments involved is
appropriate.
3. Call and Cancellation Options
An Enterprise will retire an outstanding issue of callable debt in
order to issue new debt at favorable rates. For similar reasons an
Enterprise may cancel a swap. For example, an Enterprise can cancel a
pay-fixed/receive-floating swap--which, together with discount notes,
creates a synthetic fixed-rate liability--in order to enter into a new
swap that lowers the effective cost of the synthetic liability. OFHEO
recognizes that, in general, an Enterprise will exercise its option
when the net interest cost savings on a replacement security or
contract, exceeds some threshold.
OFHEO received several comments to the ANPR that emphasized the
importance of modeling the exercise of the call option. OFHEO concurs
with these comments and, accordingly, treats callable debt in a manner
that takes into consideration the exercise of the call option. OFHEO
considered developing a financial model to value call and cancellation
options and determine when they would be exercised in the stress test.
However, the added precision of such a valuation model, as opposed to a
simpler approach, would not have a significant effect on the capital
requirement because the severe nature of the interest rate shocks
included in the stress test result in either all eligible debt being
called in a short period of time or no debt being called over the
entire period. In addition, a valuation model would add a considerable
amount of complexity to the cash flow model. Therefore, OFHEO sought to
develop an alternative approach for decisions to exercise call and
cancellation options that would provide a reasonable approximation of
the Enterprises' procedures for exercising such options without
increasing the complexity of the model.
OFHEO proposes to use, as a proxy for this threshold option value,
the spread between the coupon rate of an outstanding actual or
synthetic debt security and the Enterprise cost of funds for a new
replacement security (the call-spread). Thus, in the stress test, the
call option is exercised and the debt retired when the cost of the new
debt plus the call-spread is less than the cost of the existing debt
instrument. This methodology is often used as a simplified approach in
modeling applications and was suggested by Freddie Mac in its comments
to the ANPR. No other commenter suggested a specific approach.
To calculate an appropriate call spread, OFHEO received data from
the Enterprises on the threshold value of call options on debt, in
terms of a call-spread, over a range of reasonable times to maturity
and valuation model parameter settings. After reviewing this
information, OFHEO proposes to use a call-spread in the stress test of
50 basis points over the cost of issuing new bullet debt with the same
time to maturity as the callable debt. This call-spread provides a
reasonable debt call rule, without adding a considerable amount of
complexity to the model.
4. Counterparty Risk
The ANPR sought comment about how, if at all, OFHEO should
incorporate the effect of derivative counterparty defaults into the
stress test. The Enterprises frequently enter into derivative contracts
that, combined with various types of debt instruments (including
structured notes), create synthetic liabilities at lower cost then
actual debt with the same characteristics. Other derivative contracts
are used as macro hedges against portfolio level risks. However, all
swaps expose an Enterprise to counterparty credit risk, which is the
risk that the counterparty may default on its contractual obligation at
a time when the derivative contract has a positive market value to the
Enterprise.
[[Page 18159]]
Currently, the Enterprises limit their exposure to counterparties
by entering into swap transactions only with counterparties rated
investment grade and by requiring all counterparties to execute
collateral pledge agreements. These pledge agreements require any
counterparty currently rated or subsequently downgraded to a less than
a AAA credit rating to post collateral to the extent that net losses on
its contracts \170\ with an Enterprise exceed threshold levels. The
threshold levels vary based on the counterparty's rating. The
Enterprises do not require AAA-rated counterparties to post collateral,
but if any counterparty is downgraded, the collateral pledge agreements
subjects it to the more stringent collateral requirements of its new
lower rating. Freddie Mac, in its comments, describes additional
measures it uses to mitigate counterparty risk, which include using
contracts with close-out and netting arrangements that allow Freddie
Mac to offset losses on one contract with a particular party against
gains on another contract. Freddie Mac also described its practice of
requiring guarantees from well-capitalized parent companies and of
periodically marking each contract to market at full replacement value.
---------------------------------------------------------------------------
\170\ These losses are calculated on a mark-to-market basis,
because most derivatives involve features, such as payment streams
and options, the values of which fluctuate with changes in the yield
curve.
---------------------------------------------------------------------------
In commenting on the ANPR, Freddie Mac stated that its management
of credit risk on derivatives is such that the stress test should
specify no losses due to counterparty default. Freddie Mac suggested
that any losses would be covered adequately by the 30 percent add-on
that the 1992 Act requires for management and operations risk and by
the minimum capital standard. ACB, commenting generally on the subject
of counterparty risk, stated that where collateral is provided, the
risk of counterparty failure is remote. ACB suggested that, at most, a
straightforward ``haircut'' on ``in the money'' derivative positions
should be applied.
After consideration of these comments, OFHEO determined that
reducing the haircuts for derivative counterparty risk by 80 percent
from haircuts on other types of third party credit risk would provide
appropriate recognition for Enterprise collateral agreements. However,
OFHEO did not agree with Freddie Mac that the stress test should apply
no haircuts. There always remains the possibility that counterparties
could default on their obligations due to a sudden calamity that could
prevent collateral from being posted. Also, collateral values can
decline over time or collateral may be subject to competing claims.
Sudden business bankruptcies and decline or impairment of collateral
value would be even more likely than usual under the harsh economic
circumstances of the stress test. Accordingly, and for the same reasons
that similar haircuts are applied to mortgage credit enhancements and
non-mortgage investments, OFHEO proposes to specify losses in the
stress test due to failure of derivative counterparties.
OFHEO proposes to take into account the amount of loss due to
derivative counterparty default as follows. As illustrated in Table 29,
the stress test applies haircuts that increase linearly (by equal
amounts) each month to the net payments from derivatives with a given
counterparty over the term of the contracts with that counterparty.
That is, if the Enterprise's net swap position across all contracts
with a particular counterparty imply cash payment to the Enterprise
during a given month, that cash payment is reduced (``haircut'') by an
amount determined by the public credit rating of the counterparty and
period in which the payment is owed. The calculation is performed for
each counterparty and for each month in which a counterparty has swap
agreements with the Enterprise. The cash flows for all derivatives with
each counterparty are netted, except swaps that exchange into U.S.
dollars any currency in which Enterprise debt may be denominated.
Haircuts are applied separately to these derivatives, as explained
below.
[GRAPHIC] [TIFF OMITTED] TP13AP99.213
[[Page 18160]]
The haircuts reflect the probability that some counterparties will
be unable to meet their obligations during the stress period. Haircuts
become progressively larger as the counterparty rating decreases, with
parties rated BBB or lower and unrated parties receiving the most
severe haircut. The haircut for each rating category is cumulative
rather than additive. It increases linearly for each month of the
stress period, beginning in the first month of the stress test until
the full amount of the discount is reached in the 120th month. Table 29
reflects the size of the haircut at the end of each 12 month period
during the stress test. Rating downgrades are not modeled. Instead,
deterioration in the financial condition of counterparties due to the
stressful environment is reflected in the linear increase of the
haircuts.
The proposed approach recognizes that both Enterprises utilize
netting and close out arrangements such as those described by Freddie
Mac in its comments. If OFHEO determines that not all derivatives with
a particular counterparty are covered by a single arrangement, the
derivatives' cash flows will not all be netted together. Instead, the
stress test will group the derivatives by netting agreement and apply
haircuts separately to the net cash flow for the derivatives covered by
each agreement. For derivatives covered by no netting agreement, the
haircut would be applied on an instrument by instrument basis to any
derivatives that are ``in the money.'' In the event that any
derivatives contracts do not include standard Enterprise collateral
agreements, the haircut percentages imposed will be those in Table 27
in section III.C., Mortgage Credit Enhancements.
As mentioned above, the stress test will apply haircuts separately
to swap agreements that exchange into U.S. dollars any other currency
in which Enterprise debt may be denominated. Because these agreements
entail the Enterprise receiving payment denominated in other
currencies, which the stress test does not model, the stress test
cannot net them against more usual interest rate swaps. Neither can the
stress test net these agreements against each other, since they use
variety of currencies. Therefore, the stress test applies haircuts to
each individual contract. Because the collateral agreements and
investment ratings do not differ for the counterparties to these
agreements, the stress test applies the same counterparty haircut
percentages to them as it does for interest rate swaps. However, the
haircut is applied to the `pay' side of these contracts rather than to
the `receive' side. The effect will be a loss on each swap transaction
equal to the haircut amount. This approach recognizes that the
Enterprises use these swap agreements only to match a debt position for
which the swap agreement is a hedge.
E. Non-Mortgage Investments
In addition to mortgage investments, the Enterprises hold non-
mortgage investments \171\ that include Treasury securities, federal
funds, time deposits, Eurodollar deposits, asset-backed securities
\172\ (ABS), corporate securities, and state and municipal bonds.\173\
As of December 31, 1997, non-mortgage investments at Fannie Mae
constituted about $66.8 billion (17 percent of on-balance sheet assets)
and $13.8 billion (7.0 percent) at Freddie Mac.
---------------------------------------------------------------------------
\171\ Both OFHEO and HUD are authorized to regulate the
Enterprises' non-mortgage investment activities. OFHEO has specific
authority to ensure that the Enterprises are adequately capitalized
and operating safely (1992 Act, section 1313 (12 U.S.C. 4513)), and
HUD has general regulatory authority over the Enterprises to ensure
that the purposes of the 1992 Act are accomplished (1992 Act,
section 1321 (12 U.S.C. 4541)). While HUD's current regulations do
not contain specific provisions about the Enterprises' non-mortgage
investments, HUD issued an advance notice of proposed rulemaking
(ANPR) seeking comment about the need for it to regulate such
investments. (62 FR 68060, December 30, 1997)
\172\ ABS are similar to MBS but are backed by nonmortgage
assets, such as receivables on car loans and credit cards.
\173\ Although they are generally tax-exempt, for purposes of
the stress test, mortgage revenue bonds (MRBs) are not included in
the category State and municipal bonds. MRBs are discussed in the
section titled ``other housing assets.''
---------------------------------------------------------------------------
OFHEO considered several issues related to how the stress test
should model the cash flows associated with the Enterprises' non-
mortgage investments. The first issue concerns whether the stress test
should model cash flows from such investments at the instrument level
or at an aggregated level. Such aggregation entails grouping individual
instruments with similar terms and risk characteristics and modeling
the group as a single instrument. The proposed stress test models the
cash flows of all non-mortgage investments on an instrument-by-
instrument basis. Evaluating whether to model non-mortgage investments
on an instrument versus an aggregated level represents a trade-off
between accuracy, model complexity, and information system resources.
Instrument level modeling provides greater accuracy than modeling
aggregated investments because aggregating instruments may result in
losing information. On the other hand, instrument level modeling may
result in added complexity and require additional information system
resources. Neither of these concerns poses a significant constraint in
the case of modeling the Enterprises non-mortgage investments.
Accordingly, OFHEO believes that modeling cash flows from non-mortgage
investments is practicable and appropriate. With respect to complexity,
the model for generating cash flows uses the same types of information
for an individual instrument as it would for a synthetic instrument
representing a group of actual instruments. With respect to information
resources, OFHEO systems are capable of processing the large number of
individual investments in a reasonable amount of time.
The second issue concerns whether there should be any simplifying
assumptions in modeling the cash flows associated with non-mortgage
investments. OFHEO has decided to include the following three
simplifying assumptions which will facilitate this modeling, without
having a significant effect on the risk-based capital requirement.
First, for investments with common characteristics, the stress test
specifies one payment frequency for those instruments. Second, the
stress test standardizes prepayment speeds for ABS, i.e., how fast
principal (both scheduled principal and prepayments) is returned.
Third, the stress test will not apply different ABS prepayment speeds
in different interest rate environments, because ABS typically pay off
quickly and therefore are not significantly affected by interest rates.
In addition, the effect of specifying different prepayment speeds on
the risk-based capital requirement would not be significant, and would
add unreasonable additional complexity to the stress test.
OFHEO next considered whether the proposed stress test should, with
respect to non-mortgage investments, model their credit risk, i.e., the
risk that there will be a default on an instrument. OFHEO has
determined that it is appropriate to model such credit risk because
some issuers would be unable to meet their obligations during the
stress period. The proposed stress test ties the credit quality of non-
mortgage investments to the credit rating specified by one or more
nationally recognized public rating organizations, such as S&P or
Moody's. While public offerings usually have a single rating, they
occasionally have split ratings. In the case of split ratings, the
stress test will use the lowest rating.
The stress test first generates cash flows for a given instrument
and then reduces those cash flows by a specified percentage (i.e.,
``haircut'') based on the public rating organization. The percentage
haircut increases as the
[[Page 18161]]
rating decreases so that a highly-rated instrument will have a lower
haircut than a lower rated instrument. In the absence of a rating, the
stress test would apply the lowest rating category. The haircuts
increase linearly (i.e., in equal increments) during each month of the
stress period. Table 29 illustrates the ending haircuts in the 120th
month for each rating category. Refer to section III. C., Mortgage
Credit Enhancements for the discussion of the proposed haircuts.
[GRAPHIC] [TIFF OMITTED] TP13AP99.214
An instrument that is unrated or has a rating that is below
investment grade will receive the most severe haircut. This reflects
OFHEO's determination that it is appropriate for the stress test to
reflect high credit losses for non-mortgage investments that are more
risky than the instruments that are now included in the Enterprises'
current holdings. The Enterprises' non-mortgage investments are
currently of high quality,\174\ but the Enterprises are not statutorily
or otherwise legally required to invest solely in high quality
instruments. It is possible that an Enterprise might change its
investment practices to include non-mortgage investments with lower
credit quality.
---------------------------------------------------------------------------
\174\ For instance, in response to HUD's ANPR, Fannie Mae
commented that ``Nearly two-thirds of the [liquid investment]
portfolio is rated AAA (or the equivalent), and nearly all (98
percent) of the portfolio is rated at least A (or the equivalent).
---------------------------------------------------------------------------
F. Other Housing Assets
Other housing assets are a small category of Enterprise assets that
need to be modeled differently than retained whole loans and mortgage-
backed securities are modeled. They are primarily mortgage revenue
bonds (MRBs). They also include certain Real Estate Mortgage Investment
Conduits (REMIC) securities issued by private entities and some
interests in partnerships and joint ventures. These assets have cash
flow characteristics that vary from investment to investment, and the
data required to model cash flows precisely is not readily available.
The impact of how these assets are modeled on the stress test results
will be modest.
1. Mortgage Revenue Bonds
Mortgage revenue bonds are issued by state and local housing
authorities to raise funds for single family and multifamily mortgage
lending programs. Both single and multifamily mortgage revenue bonds
are secured by mortgage loans, reserve funds, and other credit
enhancements. Government subsidies to some multifamily projects also
provide implicit credit support. Most MRBs are tax exempt. The
Enterprises are permitted to hold up to two percent of their assets in
tax exempt securities.
OFHEO considered whether to model MRB cash flows on individually or
on an aggregated basis. The stress test models MRB cash flows bond-by-
bond. Although one modeling approach is to group securities and use
weighted average interest rates and terms to calculate future cash
flows, OFHEO determined that calculating cash flows individually is
simpler. Available computer hardware and software allow the calculation
of cash flows on many individual securities in almost the same amount
of time it takes to calculate a single cash flow using average rates
and maturities for a group. In addition, any decrease in precision that
might be introduced through pooling is avoided.
OFHEO next considered whether to calculate interest and principal
payments for the MRBs based on each security's actual structure or to
use a proxy for calculating bond payments. Interest on MRBs is paid at
the bond rate on the principal amount of the bond, but MRBs have
different schedules for principal repayment. In some MRBs, the issuer
may use principal repayments from mortgages associated with one MRB
transaction to retire bonds from another transaction. In many
transactions, issuers have substantial discretion to retire bonds
early. There is no single source of information on MRB structures, nor
is the information readily available from multiple sources.
OFHEO determined that the modeling approach used to calculate cash
flows on Ginnie Mae securities would provide a reasonable proxy for
cash flows on mortgage revenue bonds. Specifically, the bonds are
modeled as passthrough securities, with the underlying mortgage
collateral bearing a coupon 75 basis points higher than the bond
coupon. Although MRB payments are not passthroughs of mortgage loan
payments, the MRB payments are related to the mortgage payments. MRB
payments and Ginnie Mae security payments would be affected similarly
by loan terminations and by economic conditions. Further, borrowers
benefiting from MRB programs are similar to borrowers for the FHA and
VA loans that collateralize Ginnie Mae securities, and the loan
characteristics are similar. Therefore, the stress test calculates cash
flows for MRBs essentially the same way that it calculates cash flows
for Ginnie Mae securities. It amortizes the bond principal using loan
termination rates for FHA and VA loans that have the maturity of the
MRB and coupons equal to the MRB coupon plus a spread.
OFHEO considered whether to design a modeling approach specifically
for multifamily MRBs or to model cash flows for single family and
multifamily MRBs the same way. The stress test models cash flows for
multifamily MRBs as though they were single family Ginnie Mae
securities, just as it does for single family MRBs.
Modeling multifamily MRB cash flows according to the structures of
the securities is hampered by the same data problems that affect
modeling single family MRB cash flows. Therefore, the stress test needs
to use a proxy. The choice of proxy is limited. Information on
Government-insured multifamily loans is not readily available.
Enterprise multifamily MBSs are not an acceptable proxy for multifamily
MRBs, because the Enterprises' multifamily loans differ from the loans
that collateralize multifamily MRBs, and multifamily MBSs pay
differently from multifamily MRBs. Because multifamily MRBs are a very
small percentage of each Enterprise's assets and their impact on risk-
based capital is minimal, OFHEO determined that single family Ginnie
Mae securities would be used as a proxy for multifamily MRBs.
The stress test addresses the credit risk associated with MRBs by
applying the haircuts that are tied to the public
[[Page 18162]]
credit ratings of the bonds. The haircuts will be in the same amount
and will be applied in the same way as haircuts for credit enhancements
and non-mortgage investments. Currently, a sizeable majority of the
MRBs held by the Enterprises are rated AA and above.
2. Private Label REMICs
The Enterprises own a small amount of REMIC securities that are
issued by private sector entities. For most of these securities, the
information that would be necessary to calculate cash flows for the
REMIC collateral and thus for the REMIC securities is not readily
available.
As with mortgage revenue bonds, OFHEO considered whether to model
the cash flows of the REMIC securities or to model cash flows using a
proxy. The stress test uses a proxy. The stress test models cash flows
for private REMIC securities using the same modeling approach as it
uses for MRBs. The stress test amortizes the principal of the REMIC
securities using the appropriate termination rates for the coupons and
maturities.
Data that is needed to project precise cash flows is not readily
available. The costs of developing the data and reverse engineering the
REMIC securities are not warranted by any incremental refinement that
might result. Most of the REMIC securities held by the Enterprises are
rated AAA. The credit risk of the private issue REMICs will be taken
into account by applying the same haircuts as those used for MRBs.
3. Interests in Partnerships and Joint Ventures
OFHEO decided not to model gains or losses on interests in
partnerships or joint ventures, a category that totals less than $200
million, or less than 0.03 percent of Enterprise assets. These assets
carry little credit risk but generate tax losses that benefit the
Enterprises. OFHEO has determined that projecting cash flows and tax
benefits of these assets would create significant additional complexity
in the stress test, without having any material impact upon the risk-
based capital requirements. Accordingly, the stress test treats these
assets as though they remain on the balance sheet with no run-off and
no associated income. In the future, if these investments become a
larger proportion of either Enterprise's book of business, OFHEO will
reconsider how they are modeled in the stress test.
G. Commitments
The 1992 Act specifies that during the stress period the
Enterprises will purchase no additional mortgages nor issue any MBS,
except that--
[a]ny contractual commitments of the enterprise to purchase
mortgages or issue securities will be fulfilled. The characteristics
of resulting mortgage purchases, securities issued, and other
financing will be consistent with the contractual terms of such
commitments, recent experience, and the economic characteristics of
the stress period.\175\
---------------------------------------------------------------------------
\175\ 1992 Act, section 1361(a)(3)(A) (12 U.S.C. 4611(a)(3)(A)).
The 1992 Act does provide for later amendment of the rule to address
new business during the stress period, but not until after this
regulation is final. The 1992 Act requires that, within one year
after this regulation is issued, the Director of the Congressional
Budget Office and the Comptroller General of the United States shall
each submit to the Congress a study of the advisability and
appropriate form of any new business assumptions to be incorporated
in the stress test. Section 1361(a)(3)(C) (12 U.S.C. 4611(a)(3)(C)).
Subparagraph 1361(a)(3)(B) (12 U.S.C. 4611(a)(3)(B) authorizes the
Director to consider these studies and make certain new business
assumptions. However, that subparagraph does not become effective
until four years after this regulation is issued.
The term ``contractual commitments'' generally refers to binding
agreements that the Enterprises enter into with seller/servicers to
purchase mortgages or to swap mortgages for MBS. The term also refers
to agreements to sell such securities to investors. The total of
outstanding purchase or swap commitments at both Enterprises at any
point in time is generally in the tens of billions of dollars. The
following discussion describes the issues faced by OFHEO in determining
the appropriate volume and characteristics of mortgages delivered under
commitments.
1. Definition of the Term ``Commitment''
The proposed risk-based capital regulation incorporates, by
reference, the definition of ``commitment'' from OFHEO's minimum
capital regulation. OFHEO defines ``commitment'' in the minimum capital
regulation as follows:
Commitment means any contractual, legally binding agreement that
obligates an Enterprise to purchase or to securitize mortgages.\176\
---------------------------------------------------------------------------
\176\ 12 CFR 1750.2; See 61 FR 35610, July 8, 1996 (explanation
of definition).
This definition includes ``mandatory'' and ``optional''
commitments. Mandatory commitments bind the seller to deliver, and the
Enterprise to accept, a certain volume of mortgages. Optional
commitments are delivery contracts that commit the Enterprises to
purchase or swap a specified volume of loans, but do not commit the
seller to deliver any loans. The definition includes commitments that
do not specify fixed prices or volume, but otherwise legally bind an
Enterprise.
Freddie Mac, the only ANPR commenter to address the definition of
commitments, recommended that contractual commitments be defined to
include only agreements that legally bind the Enterprises to purchase
mortgages. According to Freddie Mac, ``[u]nder fundamental contract
law, an agreement is only binding if all of its key terms are included
and agreed upon.'' Freddie Mac further stated that price and volume are
two key terms and that only commitments containing this information are
legally binding contracts for the Enterprises. This comment suggests
that OFHEO should not model commitment contracts that do not contain
price and volume information (e.g., master commitments for cash
purchases).
OFHEO has found no reason to adopt a different definition for
purposes of computing risk-based capital from that used for computing
minimum capital. In both cases, the term should mean any legally
binding agreement that obligates an Enterprise to purchase or
securitize mortgages. OFHEO does not believe it necessary or
appropriate to restrict the definition of the term ``commitment'' by
reference to price, volume, and fees, because agreements may be legally
binding even when they lack specificity on all terms.\177\ It would add
unnecessary complexity to attempt to reflect the myriad details of
diverse State contract laws in the regulatory definition. Moreover, to
do so would be inadvisable in light of Congress' specific concerns
regarding the need for capital to support commitments and other off-
balance-sheet obligations. For example, in discussing the need for the
capital requirements of the 1992 Act, Congress expressed the concern
that the risk in off-balance-sheet obligations had not been captured
under prior capital standards:
---------------------------------------------------------------------------
\177\ See Restatement (Second) of Contracts Sec. 204 (1981).
The capital provisions of the GSEs' charter Acts limit their
debt to 15 times their capital unless HUD sets a higher ratio * * *
This is unsatisfactory because no capital need be held against the
GSEs' $750 billion of off balance sheet guarantees * * *\178\
---------------------------------------------------------------------------
\178\ S. Rep. No. 102-282, at 11 (1992) (referring to the
existing capital standard, which the 1992 Act repealed).
Recognizing this concern, it would be inappropriate for OFHEO to
promulgate a narrow definition that could exempt certain legally
binding commitments from the risk-based capital requirement.
Freddie Mac also recommended a definition of commitments that
excludes all optional commitments, including those containing price and
volume
[[Page 18163]]
information. Specifically, Freddie Mac suggested the following
definition:
Contractual commitment means an obligation of an Enterprise that
legally binds the Enterprise to issue securities or purchase
mortgages and legally binds a third party to purchase securities or
deliver mortgages, and that sets forth all terms of the transactions
including price, volume, and fees.
(emphasis added).
The phrase ``legally binds a third party'' would define a
commitment to include only an agreement that binds the counterparty to
deliver mortgages or to purchase securities, thus excluding optional
commitment contracts.
OFHEO disagrees with this comment and includes optional commitments
in the stress test definition. The 1992 Act is clear on this issue,
because it refers to ``commitments of the enterprise to purchase * * *
or issue'' (emphasis added) but includes no requirement that the
commitment bind others to deliver mortgages. Optional commitments
obligate the Enterprise to purchase and are optional only for the
seller. Therefore, optional commitments fall squarely within the
statutory definition.
2. Retained vs. Securitized Mortgages
The proposed regulation specifies that all loans delivered under
commitments are packaged into securities (securitized) and sold. This
specification avoids requiring OFHEO to predict business decisions by
the Enterprises that are highly judgmental and impossible to predict
accurately. OFHEO recognizes that in practice the Enterprises make day-
to-day decisions to sell or retain loans. However, the simple rule
proposed by OFHEO avoids the complexity of attempting to model such
business decisions.
ACB commented that ``[a]ny loans not presold by the GSEs should be
assumed to be retained in portfolio and carry both the credit and IRR
[interest rate risk] exposure.'' OFHEO disagrees with ACB's suggestion,
because it would add undue complexity to the stress test. At no time
are the Enterprises obligated by the terms of a commitment to retain
mortgages in portfolio. Furthermore, retaining these mortgages in
portfolio in the stress test would require OFHEO to predict how the
Enterprises would finance and hedge the interest rate risk associated
with the purchases. These predictions would increase greatly the
complexity of the stress test and introduce assumptions about future
Enterprise management, which OFHEO, as a general rule, has found
inappropriate in a ``no new business'' stress test.
For these reasons, OFHEO determined that proposing that all loans
delivered under commitments will be securitized and sold is a
reasonable, straightforward approach.
3. Modeling Delivery Percentages
The stress test will provide that, in the down-rate scenario, 100
percent of all loans that the Enterprises are obligated to accept will
be delivered and, in the up-rate scenario, 75 percent of those loans
will be delivered. As explained below, OFHEO considered the relevant
comments on this issue and found the proposed rule to be a reasonable
and practical method of estimating the volume of new mortgages that
will be delivered in the stress test.
In determining the appropriate percentage, OFHEO looked first to
the 1992 Act, which provides that commitments will be ``fulfilled.'' In
contractual parlance this term means that the parties will fulfill
their contractual obligations under these instruments. Therefore, OFHEO
decided to propose a simple rule, based upon estimates of the delivery
volumes that would be likely to occur if both parties fulfill those
obligations.
Not all mortgages that the Enterprises are obligated to accept
under commitments are actually delivered. Optional commitments obligate
the Enterprise to purchase up to a specified dollar amount of
mortgages, but do not obligate sellers to deliver any mortgages. They
can be fulfilled by both parties even though fewer than all the loans
specified in the commitment are delivered. Under a mandatory
commitment, the Enterprise is also obligated to purchase a specified
dollar value of loans, but the seller fulfills the contract either by
delivering the specified volume of loans or by paying a ``pair-off''
fee specified in the commitment agreement. These fees are a form of
liquidated damages that, under the terms of mandatory commitments, are
payable by sellers who fail to deliver the full amount of mortgages
specified in the commitments. Therefore, under either type of
commitment, less than all the stated mortgage volume may be delivered.
As mentioned above, the proposed regulation specifies that, in the
down-rate scenario of the stress test, 100 percent of loans the
Enterprises are obligated to buy or securitize will be delivered under
all types of commitments. In the up-rate scenario, 75 percent of those
loans will be delivered. This specification reflects the fact that when
interest rates decline significantly, the volume of new purchase
mortgages and mortgage refinancings generally increases. Therefore, in
the down-rate scenario, lenders should have plenty of mortgage volume
to meet or fill all commitments. In contrast, when interest rates rise
significantly, the demand for mortgages tends to fall. Therefore, in
the up-rate scenario, sellers would find it difficult to generate
enough mortgages to meet outstanding commitments. Because the proposed
regulation provides that all loan deliveries will be made in the first
three to six months of the stress period (see section III.G.4.,
Delivery Timing below), those deliveries are particularly sensitive to
short-term changes in interest rates. Thus, the steeply rising rates in
the first few months of the up-rate scenario have a significant impact
upon delivery percentages. It would be inappropriate, however, to
assume that loan deliveries would decline more than 25 percent, given
that many of the commitments are mandatory and that existing home
purchase contracts will require financing. Lenders will also have a
certain volume of outstanding loan commitments with locked rates, most
of which would close.
Figure 3 below shows that, during the most recent increase in rates
of any significance (the first half of 1994), a three month increase in
interest rates of 150 basis points led to a drop in market origination
volume of roughly 30 percent. Also, during the 12-year period shown,
market volumes never decreased over any three-month period by more than
25 to 30 percent. Because the stress test will include rate changes of
150 basis points or less in the first quarter, the data led OFHEO to
conclude that a 75 percent delivery rate would be a reasonable
specification for the up-rate scenario of the stress test.
The proposed regulation does not credit the Enterprises with income
from ``pair-off fees'' in the up-rate environment for two reasons.
First, there is no usable data on the payment of these fees or on the
percentage of deliveries under commitments. Therefore, attempting to
model these fees would require estimating, with no supporting data, the
percentages of loans to be delivered under mandatory, as opposed to
optional, commitments. Second, the fees are not always charged by the
Enterprises. Therefore, including the fees would require OFHEO to
speculate how frequently or under what circumstances the Enterprises
would impose them.
[[Page 18164]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.371
In its ANPR comments regarding delivery percentages, Freddie Mac
recommended that OFHEO develop an econometric model of delivery
percentages for commitments. This model would be based on recent
prepayment experience of each Enterprise and the prepayment rates
produced by OFHEO's default/prepayment model. The model that Freddie
Mac recommended would compute commitment delivery percentages as
follows:
1. OFHEO would determine a means of estimating the extent to which
sellers would fulfill mortgage purchase commitments by (a) delivering
mortgages or (b) paying a pair-off fee without delivering the
mortgages.
2. Then, OFHEO would determine a stress period delivery percentage
under all commitments to reflect the effect of stress period
conditions. Specifically, Freddie Mac suggested that a good
approximation of this effect would be the ratio of the sum of the
prepayment rate and the purchase-growth rate (rate of increase or
decrease in the volume of loans purchased by the Enterprises) during
the relevant portion of the stress period to the sum of the prepayment
rate and the purchase growth rate during a recent period immediately
prior to the stress period. This ratio would be multiplied by a
``baseline'' delivery percentage, which is the normal delivery
percentage during times of little interest rate fluctuation. Under this
approach, the stress test delivery percentage would be expressed as
follows:
[GRAPHIC] [TIFF OMITTED] TP13AP99.379
The stress period growth rate would be zero until such time as OFHEO
included new business assumptions in the stress test, and the stress
period delivery percentage would not be allowed to exceed 100 percent.
Freddie Mac bases its approach on two assumptions. First, the
volume of outstanding commitments at the beginning of the stress period
(i.e., the then current volume of outstanding commitments) is assumed
to be related to the volume of mortgage purchases that the Enterprises
and sellers anticipated at the time they entered into the commitments.
Second, the sellers' actual rate of deliveries during the stress period
under outstanding commitments is assumed to be closely related to
actual mortgage purchase activity during the relevant portion of the
stress period.
OFHEO agrees with these assumptions and used them to determine
appropriate stress test delivery percentages. OFHEO also agrees that an
econometric approach such as that proposed by Freddie Mac might provide
a relatively sophisticated representation of what would actually occur
under stress test conditions. However, there are insufficient data to
construct such a model of commitments at this time. Historical data
available to OFHEO do not reveal what percentages of commitments have
been delivered. The Enterprises have provided descriptions of
commitment types and made statements about their general business
practices and the length of and delivery patterns of commitments.
However, OFHEO has found available data are inadequate to associate
actual mortgage purchases with commitments. Therefore, neither of the
two steps in the Freddie Mac proposal currently is possible. There is
no source of data to determine a reasonable estimate of pair-off fee
payments or to determine a historical baseline delivery percentage.
ACB's ANPR comments suggested that a historically based dropout
factor be applied to account for failure to ``make/take delivery by
counterparties.'' The lack of historical data regarding actual delivery
percentages under commitments limits the accuracy with which such a
factor or factors can be calculated. However, OFHEO proposes an
approach consistent with the ACB comment. The stress test specifies
fixed delivery percentages for commitments in the down-rate and the up-
rate scenarios. These percentages are based on historical information,
displayed in Figure 3, about mortgage volume in the entire mortgage
market during periods when rates have risen and fallen sharply. This
information demonstrates that declining interest rates are generally
accompanied by or followed shortly by increases in the volume of
[[Page 18165]]
mortgage originations. Conversely, increasing interest rates tend to
slow originations.
4. Delivery Timing
Table 30 displays the timing of mortgage deliveries incorporated in
the stress test for both interest rate scenarios. The specified
delivery timing is consistent with the contractual terms of
commitments, the experience of the mortgage market, and the interest
rates that the 1992 Act specifies for the stress period.
[GRAPHIC] [TIFF OMITTED] TP13AP99.215
This front-loaded delivery profile in both interest rate scenarios
is consistent with the contractual terms of commitments, which usually
specify that deliveries will occur within 60 days and in most other
cases require delivery within 6 months. Also, at any point in time,
most outstanding commitments (other than those made that day) will have
only a part of the specified delivery period remaining. For these
reasons, OFHEO believes it is appropriate to project that deliveries
under commitments would drop to zero over the first three to six months
of the stress period, with half or more of those deliveries likely to
occur in the first two months.
For the same reasons that delivery percentages are higher in the
down-rate than in the up-rate environment, OFHEO believes it is
appropriate to provide for faster deliveries when interest rates are
falling than when they are rising. Mortgage origination experience
demonstrates that decreasing interest rates tend to cause significant
increases in mortgage originations. Therefore, it is reasonable to
specify that deliveries occur sooner when interest rates in the stress
test rapidly decline than when they rise.
ACB commented about delivery timing, stating that OFHEO should
assume scenarios that would be least advantageous to the Enterprises,
whether they were buying loans or selling securities. Because there are
no historical data on deliveries under commitments, the stress test
specifies delivery timing consistent with observed historical patterns
of mortgage originations. The delivery timing in the stress test is
intended to be a reasonable approximation of what would occur under the
stress test conditions specified in the 1992 Act, not necessarily what
would be least advantageous to the Enterprises.
Freddie Mac suggested two delivery timing options in its comments
on the ANPR. Freddie Mac recommended that OFHEO assume that purchases
occur uniformly over the weighted average maturity of outstanding
commitments. Alternatively, Freddie Mac suggested a formula that was
derived by assuming that commitments expire uniformly and that
purchases are uniform during the term of each commitment. Freddie Mac
described the latter approach as unnecessarily complex and unlikely to
affect the overall capital requirement associated with commitments, but
indicated it was nevertheless an acceptable means to estimate delivery
timing. OFHEO was concerned about a lack of empirical support for
either of Freddie Mac's alternative recommendations, however, and has,
therefore, chosen to propose the relatively simple delivery timing
described above.
5. Loan Mix Distribution
The type, term, LTV ratio, coupon, and geographic mix of loans
(``loan mix'') that are delivered under commitments can have a
significant impact upon associated credit losses in the stress test.
The proposed regulation provides that, with the exception of coupon
interest rates, the loan mix delivered under commitments at each
Enterprise is the same as the mix of loans securitized by each
Enterprise that were originated during the immediately preceding six-
month period. This approach reflects the view that a reasonable
indicator of the mix of loans that might be delivered in the near
future is the mix of loans delivered in the recent past. To the extent
that an Enterprise has been buying a larger or smaller percentage of
loans with a particular characteristic over the past six months, the
stress test effectively continues that mix. OFHEO's proposed approach
does not differentiate the loan mix of deliveries in the up-rate and
down-rate scenarios.
To reflect movements in stress test mortgage interest rates, the
stress test uses two different ``conventional mortgage rate'' series, a
30-year rate and a 15-year rate, described earlier in section III. B.,
Interest Rates, to determine mortgage rates on newly delivered fixed-
rate mortgages.\179\ It uses
[[Page 18166]]
the one-year CMT, along with the average margin for ARM loans
---------------------------------------------------------------------------
\179\ The stress test assumes that mortgage interest rates on
seven-year balloon mortgages are 50 basis points less than 30-year
conventional mortgage rates in the down-rate environment, and equal
to the 30-year rate in the up-rate environment.
---------------------------------------------------------------------------
originated within the past six months, to determine mortgage rates on
newly delivered ARMs.
In its ANPR comments, Freddie Mac recommended two methods of
modeling loan mix. Freddie Mac recommended that the loan mix of
mortgages delivered under commitments could be the same as the loan mix
of the Enterprises' outstanding portfolios. Alternatively, Freddie Mac
suggested that OFHEO look to historical experience and base the stress
period mix on the mix during past up-rate and down-rate environments.
Freddie Mac further commented that the mix of mortgages delivered under
outstanding commitments should not be modeled based on recent mortgage
deliveries. Its rationale was that the capital requirement associated
with commitments could vary dramatically because of one-time special
purpose transactions. Freddie Mac cited, as an example, the distorting
effects created by an Enterprise purchase of a large Cost Of Funds
Index (COFI) ARM portfolio representing 30 percent of a quarter's
purchases.
OFHEO did not adopt Freddie Mac's first suggestion because OFHEO
believed that the mix of loans in an Enterprise's overall portfolio has
only a limited relationship to the loans that will be delivered under
current commitments. An Enterprise's portfolio at any given time
contains loans obtained over many years during periods when economic
conditions may have been quite different from the conditions that will
exist at the start of the stress test. Current commitments, by
contrast, are more likely to reflect Enterprise management's efforts to
adjust the mix in its portfolio than they are to reflect the current
mix in the portfolio. For these reasons, OFHEO found the current mix of
loans at the Enterprises to be an unsatisfactory proxy for the mix of
loans to be delivered under current commitments.
Using a two-quarter (versus a one-quarter) period to compute the
loan mix addresses Freddie Mac's concern over distortions created by
occasional special purpose purchases. However, if large special purpose
purchases of unusual mortgages occur frequently, it is appropriate that
the stress test reflect some higher-than-usual risk by projecting
continuing purchases of such mortgages.
OFHEO also examined Freddie Mac's suggested alternative
methodology--basing the loan mix on the ``mix that prevailed'' during
prior up-rate and down-rate scenarios. Given the lack of historical
data regarding deliveries under commitments, there is no direct
evidence of what the experience of those deliveries has been. At best,
information might be inferred from data regarding total deliveries,
either at the Enterprises or in the market as a whole. However, OFHEO's
research has found that, although long term increases in interest rates
produce more ARMs and long term decreases produce more FRMs, short term
changes in interest rates have little discernable affect on the ratio
of ARMs to FRMs that are delivered to the Enterprises.
For these reasons, OFHEO concluded that a more detailed and complex
model based upon historical patterns of loan deliveries would be
unlikely to improve the stress test's accuracy or sensitivity to risk
or yield a significantly different result. OFHEO is confident that the
proposed approach reflects a reasonable delivery mix for the stress
test and that any fine-tuning that might result from a more complex
model would have only an incremental effect. Also, because the proposed
regulation specifies that these new loans will not be held in
portfolio, they create little interest rate risk for the Enterprises.
For all these reasons, OFHEO does not propose the type of detailed
model of loan mix contemplated in Freddie Mac's comments.
ACB also commented on loan mix, explaining that the mix of
commitments should be ``as of the actual reporting date, subject to
adjustment for any demonstrable `window dressing' practices by the
GSEs.'' ACB assumed that data were available to determine what loan mix
was specified under outstanding commitments at any point in time. As
explained above, those data are not available. OFHEO interpreted
``window dressing,'' to mean attempts that an Enterprise might make to
alter temporarily the loan mix in its commitments just prior to the
beginning of a particular quarter. OFHEO believes that the proposed
approach, which looks to the mix of loans actually delivered over the
last two quarters, addresses ACB's concern that an Enterprise might
engage in ``window dressing.''
6. ``No New Business'' Rule
World Savings commented in response to the ANPR that the stress
test model should reflect ongoing business, not a wind down scenario.
The comment stated that the assumption of no new business except for
fulfillment of contractual commitments is ``fundamentally flawed,''
because it assumes the Enterprises will be prescient about the
magnitude of the financial stress. World Savings commented that this
assumption causes the test to underestimate the Enterprises' need for
capital, because it causes their portfolios to shrink unrealistically.
By contrast, a comment by Professor Yezer of George Washington
University advocated placing limits on the size of the Enterprises'
portfolios in the stress test. He concluded that ``one needs a model of
[Enterprise] response to stress that makes sense in terms of modern
financial theory of investment, not passive reaction to adverse changes
as contemplated in the proposed rule.''
Both of these comments suggest an alternative approach to new
business that cannot be addressed at this time because the approach in
the regulation is mandated by section 1361(a)(3) of the 1992 Act.\180\
That section requires that the initial risk-based capital regulation
assume that the Enterprises take on no new business other than
deliveries under existing commitments. After the issuance of the
regulation, the 1992 Act requires studies by the Congressional Budget
Office and the Comptroller General of the United States of the
advisability and appropriate form of any new business assumptions to be
incorporated in the regulation. Only after completion of those studies
and their submission to the Congress may the Director, after
considering them, propose amendments to the regulation that would
incorporate new business assumptions during the stress period.\181\
---------------------------------------------------------------------------
\180\ 1992 Act, section 1361(a)(3) (12 U.S.C. 4611(a)(3)).
\181\ 1992 Act, section 1361(a)(3)(B)-(D) (12 U.S.C.
4611(a)(3)(B)-(D)).
---------------------------------------------------------------------------
H. New Debt and Investment Rules
During the stress period, an Enterprise invests and borrows, as
needed, based on net cash flows. The stress test projects cash inflows
and outflows for each month of the stress period. To the extent cash
inflows exceed cash outflows in any month, the stress test must specify
how an Enterprise employs the excess funds. Conversely, to the extent
that cash outflows exceed cash inflows in any month, the stress test
must specify how an Enterprise obtains the funds to cover the cash
deficit.
The 1992 Act provides no specific guidance for new debt issuance or
new investments during the stress test. OFHEO sought new debt and new
investment rules that would alter as little as possible the credit and
interest rate exposures of an Enterprise generated by its initial
asset, liability, and derivative positions.
The proposed approach provides that all new debts and investments
are short-
[[Page 18167]]
term instruments. More specifically, OFHEO proposes that the
Enterprises fund all monthly net cash outflows during the stress test
by issuing six-month discount notes. OFHEO also proposes that excess
funds will be invested at the six-month Treasury bill rate in
instruments that mature one month later.
1. Rationale for New Debt and New Investment Rules
The purpose of a ``no new business'' stress test is to subject an
Enterprise's business at the beginning of the stress period to adverse
conditions, without introducing during the stress period any business
responses to deteriorating business conditions that would tend to
increase or decrease risk. Consistent with this purpose, the proposed
new debt and investment rules are designed to project the effects
during the stress period of specific stressful circumstances on the
Enterprises, given the risks embodied in their business positions at
the start of the stress test, while minimizing the introduction of any
new risks.
Accordingly, the stress test uses simple rules for the issuance of
debt or the investment of liquidity. OFHEO intentionally does not
propose to predict what asset-liability management decisions an
Enterprise might make, predictions that would be difficult in any
event.\182\
---------------------------------------------------------------------------
\182\ In a stress test that incorporates new business, the
context would be different. Should OFHEO choose to incorporate new
business in a later regulation, a different approach to asset-
liability management during the stress period could be appropriate.
See 1992 Act, section 1361(a)(3)(C) (12 U.S.C. 4611(a)(3)(C)).
---------------------------------------------------------------------------
The hazards of predicting the response of financial institutions to
stressful conditions are well illustrated by the behavior of the
thrifts during their financial crisis in the 1980s. While some
institutions sought to limit or reduce their risks in that difficult
environment, others made choices that greatly increased risk, in effect
gambling that a fortunate turn of events would be their best chance of
financial salvation. These choices largely determined the fate of the
institutions. Similarly, incorporating activities that project the
Enterprises's responses to the duration or severity of economic
conditions during the early part of the stress period, while these
conditions are deteriorating rapidly, could profoundly affect the
Enterprises' financial performance in the stress period.
For these reasons, the stress test makes no provision for an
Enterprise to rebalance its portfolio as its asset and liability
positions evolve during the stress test. The Enterprises are exposed to
interest rate risk principally because changes in interest rates cause
changes in the market (and economic) values of their long-term, fixed-
rate assets and liabilities, and of their derivative contracts. These
changes in value are reflected in subsequent accounting statements of
earnings and net worth.
If an Enterprise's asset, liability, and derivatives positions are
well matched, the effects will be minimal. But if, for example, an
Enterprise were to fund long-term, fixed-rate mortgages with short-term
debt, then an increase in market yields would cause the value of the
mortgages to fall, but the value of the short-term debt would be little
changed. In subsequent periods, interest income on the mortgages would
be unaffected, but interest expenses would be higher because new debt
would need to be issued at the new higher interest rate. Earnings and
equity would suffer. Conversely, a fall in market yields would increase
the value of the mortgages, and that higher value would be reflected in
subsequent earnings and equity gains. If an Enterprise were to fund
short-term assets with long-term, fixed-rate debt, its debt would
change in value, but its assets would not, producing the opposite
effect.
If changes in interest rates continue over a period of time, then a
decision to issue long-term debt or purchase long-term assets in the
middle of the stress period would create a new source of changes in
value over the remainder of the period. The effects of the change in
interest rates on future earnings and equity would then reflect the
changes in value of both the original positions and the new long-term
debt or assets.
In the proposed stress test, interest rates change substantially
and continuously during the first year of the stress period and then
are constant in the last nine years. If an Enterprise were projected to
issue long-term debt or purchase long-term assets during the first
year, the new investments would change in value during the remainder of
the year and affect subsequent earnings and equity. Such an approach
would distort the stress test's evaluation of starting risk positions.
The proposed rule avoids these problems by making all new debt and
investment short-term instruments. Investments are made in Treasury
bills to avoid introducing credit risk; new debts are in the form of
discount notes. Maturities of six-months were chosen as a
representative short term.\183\
---------------------------------------------------------------------------
\183\ Recurring patterns in cash flows can cause an Enterprise
to hold substantial volumes of new six-month investments at the same
time that it has substantial volumes of new six-month debt
outstanding. This creates an unnecessary balance sheet expansion. A
more realistic solution would be to assume that maturities of new
debts and investments were spread across a variety of terms less
than one year. OFHEO proposes to approximate that result by assuming
that any outstanding new six-month investments are redeemed at par
at the end of each month.
---------------------------------------------------------------------------
2. Analysis of ANPR Comments
In the ANPR, OFHEO posed several questions related to new debt and
investments during the stress period. HUD and ACB recommended in their
comments that OFHEO develop an econometric model of Enterprise funding
decisions. OFHEO believes, however, that it would be inappropriate to
build such a model. The factors that would have to be incorporated into
such a model would require OFHEO to make complex judgments about the
decisions an Enterprise's management might make in response to future
economic conditions. HUD's comment that ``OFHEO may be able to base
modeling of GSE liability management * * * on presumptions concerning
how GSEs would formulate and exercise broad financial management
objectives during a winddown'' would require similar judgments. ACB
also commented that ``excess cash balances should be assumed to be
deployed to minimize remaining interest rate risk exposure since the
costs of such a hedging strategy are zero.'' OFHEO determined that this
approach could change the risk profile of an Enterprise during the
course of stress period and is, therefore, inappropriate for the stress
test.
Freddie Mac also addressed the question of new debt in the stress
test. Freddie Mac proposed that OFHEO assume the Enterprises would
generally adhere to their respective asset and liability management
principles in a stress test environment. More specifically, the
Enterprises would rebalance their portfolios of assets and liabilities
during the stress period, in an attempt to maintain a specific
relationship between the net effective maturity and net callability of
assets and liabilities. Freddie Mac further suggested that OFHEO should
use a simple rule that includes this concept for the issuance of new
debt in the stress test. As a possible rule, Freddie Mac offered the
following example: 30 percent short-term and 70 percent long-term debt
in the up-rate scenario and 70 percent short-term and 30 percent long-
term debt in the down-rate scenario. The intent of the stress test is,
however, to test the ability of an Enterprise's initial asset and
liability mix to survive stressful conditions. Therefore, OFHEO
preferred an approach that did not
[[Page 18168]]
actively alter the consequences of the interest rate risk exposure
inherent in the Enterprises' business at the beginning of the stress
period.
At HUD's suggestion in its comments on the ANPR, OFHEO reviewed the
role of new debt in the wind down scenarios described in HUD's 1987
Report to Congress on FNMA, issued on September 27, 1989. Although
OFHEO agrees with HUD that there is a close connection between
investing cash, hedging activities, and liabilities, OFHEO believes
that the purpose of the ``no new business'' stress test is to project
the results of existing risk positions in stressful environments. This
approach differs significantly from HUD's 1987 wind down scenarios,
which were designed to project Fannie Mae's performance during an
intentional wind down of Fannie Mae's mortgage portfolio in preparation
for a hypothetical privatization of that Enterprise.
I. Operating Expenses
Operating expenses include non-interest costs, such as those
related to an Enterprise's salaries and benefits, professional
services, property, and equipment. The operating expenses of each
Enterprise comprise a relatively small portion of their overall
expenses. For instance, in 1997, Freddie Mac's interest-related
expenses were $10.6 billion, while its operating expenses were $495
million. Similarly, Fannie Mae's interest-related expenses were $22.4
billion, while its operating expenses were $636 million that year.
The 1992 Act is silent on how operating expenses should be treated
in the stress test. Nevertheless, the legislative history states that
the Director should exercise discretion about variables such as the
Enterprises' operating expenses, provided that they are ``reasonable
and to the extent possible based on historical data.'' \184\ In
addition, the stress test's treatment of operating expenses is guided
by the 1992 Act's ``no new business'' requirement.\185\ That provision
requires OFHEO to project the income and expenses associated with the
existing business positions of the Enterprises over a ten-year period.
The purpose of the ``no new business'' requirement is for the stress
test to capture the risks of an Enterprise's existing assets,
liabilities, and off-balance sheet obligations as of the beginning of
the stress period. It is not intended to represent any combination of
events that might occur in the actual course of an Enterprise's
business activities.
---------------------------------------------------------------------------
\184\ H.R. Rep. No. 102-206, at 65 (1991).
\185\ 1992 Act, section 1361(a)(3)(A) (12 U.S.C. 4611(a)(3)(A)).
---------------------------------------------------------------------------
In the proposed regulation, operating expenses decline during the
stress period in direct proportion to the decline in the volume of each
Enterprise's total mortgage portfolio (i.e., the sum of the outstanding
principal balance of its retained and sold mortgage portfolios). The
stress test first projects how an Enterprise's mortgage portfolio
decreases during the stress period on a monthly basis. After
determining the percent of these assets that remain at the end of any
month during the ten-year stress period, OFHEO simulates the reduced
operating expenses in each month by multiplying this percent by one-
third of the amount of the Enterprise's operating expenses in the
quarter immediately preceding the start of the stress test. This
computation is used to determine the Enterprises' operating expenses
for each month of the stress period. As described in more detail in
this section below, under this approach, the expense reduction pattern
for the up-rate scenario will differ from the down-rate scenario, and
the pattern within each scenario will vary depending on changes in the
characteristics of an Enterprise's total mortgage portfolio.
In the ANPR, OFHEO raised several questions about how the stress
test should model operating expenses. These issues are considered
below.
OFHEO first considered whether there should be any reduction in
operating expenses during the stress period. The stress test should
include such a reduction because many of the Enterprises' operating
expenses are tied to the size of their mortgage portfolios. Both
commenters on this issue, Freddie Mac and ACB, supported this view.
OFHEO next considered whether there should be a variable or
straightline reduction in operating expenses. OFHEO determined that a
variable reduction pattern would be more appropriate. The underlying
characteristics of mortgages held or guaranteed by an Enterprise or the
interest rate conditions of the stress period would substantially
affect the rate of reduction in outstanding mortgage balances. Because
a large portion of expenses are directly tied to outstanding loan
balances, a variable reduction based on those balance patterns will
better correspond with the cost reductions that would occur under the
stress test scenarios.
Notwithstanding this general approach, OFHEO notes that expenses in
some categories are not closely tied to current loan balances. These
expenses might be expected to change at different rates from loan
balances in a stressful no-new-business environment. As Freddie Mac
commented in response to the ANPR, a large portion of its operating
expenses are associated with either new business or long-term research
and development, including product and systems development, and so
might be reduced more dramatically under a no-new-business assumption.
Conversely, Freddie Mac stated that some other operating costs that are
associated with ongoing costs of managing the mortgage portfolio are
relatively fixed, i.e., they are independent of the size of the
portfolio. On balance, tying expenses to loan balances will produce a
reasonable approximation of an Enterprise's costs in the stress test
scenarios.
The proposed approach to modeling operating expenses differs from
the recommendations made by ACB and Freddie Mac. Rather than a variable
approach, these commenters favored a model applying a straightline
reduction in operating expenses. Freddie Mac commented that a
straightline approximation is sufficient, because the resulting capital
requirement should depend primarily on the present value of the
operating expenses and not on the exact timing of those expenses.
However, OFHEO believes it is appropriate to adopt an approach that
more precisely takes timing into consideration, because the timing of
expenses affects an Enterprise's performance during the stress test and
the resulting risk-based capital requirement. Furthermore, a
straightline approach still requires a basis on which to determine the
rate of expense reduction. The proposed approach simultaneously takes
timing into account and determines the overall rate of reduction.
The next issue concerned whether the model should reflect decisions
that might be made by an Enterprise if it was intentionally winding
down its business. On that issue, HUD recommended two alternative
approaches: either that OFHEO model the behavior of an Enterprise on
issues such as liability management, dividend policy, and operational
management as if it were aware that a wind down is in effect, or that
OFHEO proceed in a ``more formalistic fashion,'' i.e., without regard
to whether they did or did not know. OFHEO analyzed this issue, not
only within the context of operating expenses, but also as it relates
to the underlying concepts of the stress test and many of its
components. OFHEO determined that it would be inconsistent with the
1992 Act and the overall purposes of the stress test for the
[[Page 18169]]
model to attempt to reflect decisions that would be made by an
Enterprise that was intentionally winding down its operations. Instead,
the stress test applies the alternative approach discussed by HUD in
which an Enterprise would not know that a wind down was in effect. As
discussed earlier, this approach is appropriate because the stress test
is intended to capture the actual risks of an Enterprise's existing
business as of the beginning of the stress period rather than events
that might occur during the actual course of its business.
OFHEO next considered whether it is appropriate to treat categories
of operating expenses differently. OFHEO has determined that
disaggregating the operating expenses into several categories would add
needless complexity without providing any significant corresponding
benefit to ensuring an Enterprise's capital adequacy. While some
expense categories might reasonably be assumed to decline faster than
the mortgage portfolio, some others might decline more slowly, and some
might be expected to increase. OFHEO agrees with ACB and Freddie Mac
that since operating expenses constitute a relatively small portion of
an Enterprise's overall costs, they should not be subject to
complicated modeling. Accordingly, OFHEO proposes to consider operating
expenses in a single category rather than disaggregating them into
distinct categories.
Finally, OFHEO considered whether the operating expenses of each
Enterprise should be modeled in the same manner. Freddie Mac
recommended that instead of distinguishing between the Enterprises, the
stress test should reduce operating expenses of each Enterprise in the
same manner. Freddie Mac stated that any attempt to make fine
distinctions between how each Enterprise might actually manage its
operating expenses during the stress period could lead to extensive
analysis that ought to have little affect on the overall capital
requirement but, could increase the danger of different capital
treatment for each Enterprise based on differences in accounting
treatment of expenses.
OFHEO agrees with Freddie Mac's recommendation not to distinguish
between the Enterprises with respect to modeling operating expenses. A
fundamental concept of the risk-based capital requirement is that the
stress test establish a single set of rules that apply equally to both
Enterprises. It would be inappropriate to establish a different stress
test for each Enterprise. As a result, differences in operating
expenses during the stress test between the Enterprises will reflect
only differences in initial expense levels and mortgage portfolio
composition, not any projected behavioral differences.
J. Dividends and Other Capital Distributions
1. Introduction
The definition of a ``capital distribution'' in the 1992 Act
includes the payment of common stock dividends, preferred stock
dividends, and the repurchase or retirement of shares of stock.\186\ In
recent years, both Enterprises have consistently paid significant
amounts of dividends and have repurchased significant amounts of common
stock.
---------------------------------------------------------------------------
\186\ 1992 Act, section 1303(2)(A) (12 U.S.C. 4502)(A)). The
notable exception is the repurchase of shares for employee stock
ownership programs under section 401 of the Internal Revenue Service
Code of 1986.
---------------------------------------------------------------------------
The 1992 Act directs OFHEO to consider dividends in the stress
test. When an Enterprise makes a capital distribution and the amount of
that distribution, however, are not specified in the 1992 Act. The only
requirement is that dividends should be consistent with the stress test
environment.\187\ Because capital distributions decrease equity, the
more distributions an Enterprise makes during the stress test period
(or during a real-life stressful environment), the more likely that an
Enterprise will fail to meet its risk-based capital requirement.
---------------------------------------------------------------------------
\187\ 1992 Act, section 1361(b)(2) (12 U.S.C. 4611(b)(2)).
``Characteristics of the stress period other than those specifically
set forth in subsection (a), such as prepayment experience and
dividend policies, will be those determined by the Director, on the
basis of available information, to be most consistent with the
stress period.''
---------------------------------------------------------------------------
2. Statutory Provisions
The 1992 Act and the Charter Acts determine the authority of the
Enterprises to make capital distributions.\188\ Under these statutes,
an Enterprise may make a capital distribution without restriction when
the Enterprise would remain adequately capitalized following the
distribution.\189\ In all other circumstances, a capital distribution
is prohibited outright or requires the approval from the Director of
OFHEO.
---------------------------------------------------------------------------
\188\ Fannie Mae's Charter Act and Freddie Mac's Corporation Act
collectively are referred to as the ``Charter Acts.''
\189\ In general, an Enterprise is considered ``adequately
capitalized'' when it meets both the risk-based and minimum capital
levels. It is ``undercapitalized'' when it does not meet the risk-
based capital level, but does meet the minimum capital level. It is
``significantly undercapitalized'' when it does not meet either the
risk-based capital level or the minimum capital level, but does meet
the critical capital level. See section 1364 of the 1992 Act (12
U.S.C. 4614), and section 303(c)(1) of the Charter Act and section
303(b)(1) of the Corporation Act.
---------------------------------------------------------------------------
Prior approval by the Director is required when an Enterprise is
undercapitalized or if a capital distribution would cause the
Enterprise to be undercapitalized.\190\ The legislative history of this
requirement makes clear that, while approval in these circumstances can
be granted, such approval ``should be the exception and not the rule.''
\191\ The Director's prior approval also is required when an Enterprise
is significantly undercapitalized; however, the 1992 Act places
conditions on the granting of such approval. In those circumstances,
the Director may only approve a distribution if the Director determines
that it will: (1) Enhance the Enterprise's ability to meet its capital
requirements, (2) contribute to the Enterprise's long term safety and
soundness, or (3) is otherwise in the public interest.\192\ No approval
may be granted for a distribution that would cause the Enterprise to be
significantly undercapitalized or critically undercapitalized.\193\
---------------------------------------------------------------------------
\190\ Section 303(c)(2) of the Charter Act and section 303(b)(2)
of the Corporation Act.
\191\ S. Rep. No. 102-282, at 24 (1992).
\192\ 1992 Act, section 1366(a)(2) (12 U.S.C. 4616(a)(2)).
\193\ 1992 Act, sections 1365(a)(2); 1366(a)(2)(A) (12 U.S.C.
4615(a)(2); 4616(a)(2)(A)).
---------------------------------------------------------------------------
This statutory structure draws a clear distinction between an
Enterprise that fails to meet its risk-based requirement and one that
fails to meet its minimum capital requirement. When an Enterprise fails
to meet the risk-based capital requirement, the Director has full
discretion to grant or deny approval for a capital distribution.
However, when an Enterprise fails to meet the minimum capital
requirement, the Director's discretion is limited. Moreover, the
Director is prohibited from approving a distribution that would cause
the Enterprise to fail to meet the minimum capital requirement.
3. Proposed Approach
The proposed regulation provides that during the stress period:
When paid, dividends are paid at rates consistent with
historical experience;
Dividends are paid on common stock when the Enterprise
meets the risk-based capital requirement and the minimum capital
requirement;
Dividends are paid on preferred stock when the Enterprise
meets the minimum capital requirement; and
No dividends are paid when the Enterprise does not meet or
would not
[[Page 18170]]
after payment of the dividend meet the minimum capital requirement.
In making this proposal, OFHEO emphasizes that there are
significant differences between establishing a dividend payment policy
for the risk-based capital requirement and acting on a dividend
approval request from an Enterprise that is no longer adequately
capitalized. Accordingly, provisions of the stress test which provide
for the payment of dividends by an undercapitalized Enterprise in some
circumstances and not others should not be interpreted as an indication
of how OFHEO will act on any specific dividend approval request. In
practice, OFHEO will evaluate any request for approval of a dividend
payment on the basis of a case-by-case analysis of all the relevant
facts and circumstances.
a. Preferred Stock
Under the proposed regulation, dividends are paid on preferred
stock during the stress period when the Enterprise meets its estimated
minimum capital requirement. Preferred stock dividends are based on the
coupon rates of the issues outstanding. The coupon rates for any issue
of variable rate preferred stock is calculated using projections of the
appropriate index rate.
To determine whether the Enterprise meets the minimum capital
requirement, the stress test computes the minimum capital level each
month by applying the appropriate leverage ratios to all assets (2.50
percent) and off-balance sheet obligations (0.45 percent). OFHEO notes
that interest rate and other off-balance sheet contracts also affect
the minimum capital number.\194\ However, incorporating these features
in the calculation would require OFHEO to compute the credit equivalent
amount of interest rate and foreign exchange contracts, which would add
unnecessary complexity but provide little corresponding benefit.
Accordingly, for purposes of determining dividend payouts in the stress
test, OFHEO believes that the approach described above provides a
reasonable approximation of the minimum capital calculation.
---------------------------------------------------------------------------
\194\ 12 CFR 1750.4.
---------------------------------------------------------------------------
As noted above, preferred stock dividends are paid in some
circumstances in which common stock dividends are not paid. The stress
test includes this distinction based on the recognition that when a
corporation issues preferred stock, it is making a higher level of
commitment to those investors than when it issues common stock.
Preferred stockholders have a first claim on distributions. Therefore,
failure to pay dividends on both classes of stock likely would have
greater repercussions on an Enterprise's funding costs and ability to
attract new equity capital than would a failure to pay common stock
dividends while preferred stock dividends were maintained. Accordingly,
when an Enterprise is classified as undercapitalized, the stress test
pays preferred stock dividends, but not common stock dividends.
b. Common Stock
Under the proposed regulation, dividends are paid on common stock
during the first four quarters of the stress period. The stress test
specifies that common stock dividends cease after that, reflecting the
strong likelihood that an Enterprise would not meet the risk-based
capital requirement during the final nine years of the stress period.
The rate at which dividends are paid is based on the trend in the
Enterprise's earnings. If earnings are positive and increasing,
dividends are paid based at the same dividend payout ratio as the
average payout ratio of the four quarters preceding the stress test.
Otherwise, dividends are paid based at the preceding quarter's dollar
amount of dividends per share. Dividends would be cut off before the
end of the first year if an Enterprise failed to meet its estimated
minimum capital requirement.
OFHEO believes this rule is based on a reasonable representation of
when an Enterprise will no longer be adequately capitalized. The
conditions of the stress test are sufficiently stressful to assure that
the Enterprise would be undercapitalized by the end of the first year
of the stress period. By that time, an Enterprise's portfolio would
have been subjected to very large interest rate increases or decreases.
If, at that point, it was subjected to those same large increases,
i.e., a total of up to 1200 basis points over two years, it is
reasonable to assume that the Enterprise would be undercapitalized. The
Enterprise would have to withstand more severe credit losses because
the hypothetical stress tests would also compound declines in house
prices associated with the actual stress test. Estimating with greater
accuracy whether an Enterprise would meet its risk-based capital
requirement at any time during the stress period is inherently
difficult. This would require simulating a series of hypothetical ten-
year stress tests, the last of which would involve generating cash
flows extending ten years beyond the end of the actual stress period.
This would add great technical complexity to the stress test without
providing any meaningful benefit.
c. Other Types of Capital Distributions
The proposed regulation does not provide for any other types of
capital distributions, such as repurchases of common stock, or
redemption of preferred stock. Although the Enterprises have both
repurchased a significant number of shares of their own common stock in
the past several years, the stock buybacks were irregular events based
on the current share price, expected return on potential investments,
and the profitability of each Enterprise. The Enterprises have made no
firm commitment to investors to continue share repurchases.
Furthermore, OFHEO believes that the stress test environment would not
be conducive to share repurchases.
4. Analysis of ANPR Comments
In response to questions in the ANPR, Freddie Mac emphasized that
any assumptions that OFHEO makes regarding dividend payments must be
consistent with the 1992 Act, particularly the provisions related to
how capital classifications affect dividend payments. With regard to
preferred stock dividends, Freddie Mac recommended that OFHEO assume
that an Enterprise pays dividends on such stock so long as it satisfies
its minimum capital requirement and discontinues preferred dividends
thereafter. With regard to common stock dividends, Freddie Mac
recommended that OFHEO assume that an Enterprise pays a constant
dividend payout ratio on common stock until earnings become negative,
at which time common stock dividends would be discontinued.
The proposed regulation, which ties dividend payouts to capital
classifications, is consistent with the 1992 Act and is generally
consistent with Freddie Mac's recommendations. More specifically, OFHEO
agrees with Freddie Mac's recommended approach for paying preferred
stock dividends until an Enterprise's capital falls below the minimum
level. OFHEO believes this treatment of preferred stock dividends
properly reflects the high level of commitment of the Enterprises to
investors in their preferred stock.
In addition, eliminating common stock dividends after an Enterprise
becomes undercapitalized is roughly equivalent to Freddie Mac's
recommendation to cut off common stock dividends when an Enterprise's
earnings turn negative. However, while Freddie Mac would reduce
dividends proportionately if earnings decline, the proposed regulation
provides for the
[[Page 18171]]
payment of a constant dollar amount. OFHEO believes the payout rule in
the stress test appropriately reflects the current dividend payout
history of the Enterprises. Both Enterprises have made fairly strong
commitments to investors regarding dividend payouts, and have been slow
to lower their dividend payments in the face of declines in earnings.
ACB recommended that dividends be suspended immediately in the
stress test, since the Enterprises are assumed to be in a wind down and
shareholders would be strictly residual claimants. ACB's recommendation
to suspend all dividends immediately is not consistent with the
apparent intent of the 1992 Act, which specifically mentions dividend
policies and directs OFHEO to consider dividend policies that would be
``most consistent with the stress period.''\195\ As discussed above,
OFHEO believes that the proposed capital distribution rule is
consistent with the stress test period. Furthermore, the stress test
would fail to incorporate a likely source of capital depletion that
would affect an Enterprise in a real-life stressful environment if all
capital distributions were eliminated during the entire stress test
period.
---------------------------------------------------------------------------
\195\ 1992 Act, section 1361(b)(2) (12 U.S.C. 4611(b)(2)).
---------------------------------------------------------------------------
ACB's comment that shareholders would be strictly residual
claimants, which implies that the stress test is a liquidation
situation, is not consistent with the concepts underlying the stress
test. A wind down or ``no new business'' stress test is not the
equivalent of a liquidation. Rather, it is a test of how much capital
an Enterprise would need to survive.
K. Other Off-Balance Sheet Guarantees
In addition to guaranteeing mortgage-backed securities they issue
as part of their mainline business, the Enterprises occasionally
guarantee other securities. Such guarantees are referred to as ``other
off-balance sheet (OBS) guarantees.'' Examples of other OBS guarantees
include guarantees of tax-exempt multifamily housing bonds issued by
state and local government agencies, Enterprise-issued whole loan REMIC
securities to security, and private label (non-GSE-or GNMA-issued)
REMIC securities. In general, an Enterprise's guarantee is protected by
other credit enhancements, including reserve funds, insurance
arrangements, and/or subordinated security tranches.
For the following reasons it is not now feasible to simulate the
detailed financial impact on an Enterprise of other OBS guarantees over
the 120 months of the stress period. First, the mortgage collateral for
such securities is often dissimilar from the Enterprise's mortgages on
which the stress test's mortgage performance models are based. Second,
current data on the status of the underlying collateral is difficult to
obtain. Third, the structures of the securities and the nature of
credit enhancements vary, requiring the individual modeling of each
guaranteed security, which would, at this time, require an inordinate
amount of resources.
The stress test utilizes a proxy for the detailed modeling of the
impact of other OBS guarantees on the amount of starting capital that
an Enterprise would need to just maintain positive capital during the
stress period. The proxy treatment consists of multiplying the
outstanding balance of all other guarantees at the beginning of the
stress period by .0045, and adding the result to the amount of starting
capital calculated for all other aspects of an Enterprise's operations.
The multiple .0045 corresponds to the minimum capital requirement
associated with these other OBS guarantees.
L. Calculation of the Risk-Based Capital Requirement
1. Proposed Approach to Calculating Capital
The 1992 Act requires an Enterprise to meet the risk-based capital
requirement. To determine this requirement, the statute establishes a
two-step process. The first step is to determine the amount of capital
that an Enterprise needs to just maintain positive capital during a
ten-year period of economic stress. The second step is to increase that
amount of capital by another 30 percent to capture management and
operations risk.
OFHEO proposes to use a present value approach to calculate the
capital that an Enterprise needs to just maintain positive capital
during the stress test. Once the stress test has projected the capital
of an Enterprise at the end of every month in the stress period, the
capital calculation process discounts the monthly capital balances back
to the start date of the stress period. The Enterprise's starting
capital is then adjusted by subtracting the lowest of the discounted
capital balances to account for the smallest capital excess or largest
deficit (subtracting a negative number in the case of a deficit). The
discount factor used to discount a monthly capital balance is based on
after-tax borrowing or investing yields (as appropriate) for that month
and all previous months during the stress period.
After the stress test ascertains the amount of capital necessary to
just maintain positive capital during the stress test, it then
multiplies that amount by 1.3 to arrive at the risk-based capital
requirement.
2. Justification for Using a Present Value Approach
The 1992 Act requires OFHEO to determine the amount of capital that
is sufficient for an Enterprise to just maintain positive capital
during the ten-year stress period. However, when an Enterprise has more
(or less) capital than it needs to just maintain positive capital, the
law does not specify the procedure for calculating how much capital it
would need to just maintain positive capital.
In analyzing the best method to calculate capital during the ten-
year stress period, OFHEO considered two approaches: (a) the present
value approach, described above, and (b) an ``iterative approach'' in
which the stress test would be run multiple times with hypothetical
adjustments made to each Enterprise's balance sheet prior to each run.
The present value approach more efficiently produces results comparable
to the iterative approach. Both approaches recognize that a dollar
today is worth significantly more than a dollar ten years from now,
because the dollar can be invested so as to return more in a later
year.
Under the iterative approach, the capital calculation process
begins by running the stress test on the basis of an Enterprise's
actual assets, liabilities, net worth, and off-balance sheet items as
of a given date. The first stress test run would be used to identify
the lowest capital balance that the Enterprise has during the stress
period. Then, based on that result, adjustments would be made to the
starting capital and the assets and/or liabilities on the Enterprise's
balance sheet. The goal of these adjustments is to construct a starting
position book of business that, when subject to the stress test, will
result in the Enterprise just maintaining positive capital during the
stress test. If a run results in the Enterprise's capital reaching a
minimum point greater than zero, OFHEO would reduce the starting
capital in order to move the minimum point down toward zero in the next
run. If a run resulted in the Enterprise's capital reaching a minimum
point less than zero, then OFHEO would increase the starting capital in
order to move the minimum point up toward zero in the next run. If the
second run did not achieve the desired result, successive runs would be
made following further
[[Page 18172]]
adjustments to the starting position balances.
OFHEO is proposing the present value approach rather than the
iterative one based on the following considerations. The present value
approach is comparatively simple and easy. It will not require explicit
changes to an Enterprise's actual assets, liabilities, net worth, and
off-balance sheet items as they exist at the start of the stress test,
and it achieves results comparable to the iterative approach. It
achieves these results because the discount factors used in the present
value calculations, which calculate the surplus or deficit of starting
capital, are consistent with the effects during the stress period of
the balance sheet adjustments required by the iterative approach. The
discount factors reflect the yields on additional debt or investments
offsetting necessary changes in starting capital. For example, consider
a scenario in which an Enterprise holds more starting capital than
necessary to maintain positive capital throughout the stress period.
Balance sheet adjustments made for the final iteration would likely
involve substituting for the surplus starting capital an equal amount
of debt. Discounting the appropriate monthly capital balance during the
stress period, using stress period yields, results in a comparable
amount.
Based on these considerations, the present value approach would be
a more appropriate methodology for carrying out the purposes of the
statute. The iterative approach would add needless complexity and
require OFHEO to make changes to the balance sheets of the Enterprises.
Each iterative run, would be based on hypothetical representations of
the Enterprise's position. The present value approach eliminates the
need for these artificial adjustments and the unwarranted complexity
that the iterative approach's adjustment process would entail.
Under the present value approach, it is necessary to determine the
appropriate monthly discount rates. In determining the monthly rates,
OFHEO sought a set of discount rates that would reflect the time value
of money to an Enterprise during the stress period. Accordingly, the
discount rates applied in the stress test are computed as an after-tax
rate. Such an after-tax rate reflects the fact that any borrowing
necessary to fund an Enterprise's business activities would be
deductible for income tax purposes. Conversely, any additional earnings
would be subject to income taxes.
These discount rates are intended to reflect the fact that interest
rates will differ dramatically between the rising and falling rate
scenarios and at given times in each scenario. When an Enterprise is
borrowing new funds during the stress period, the marginal effect that
a change in its cash position in one month will have on its equity in a
subsequent month will be reflected by its after-tax cost of borrowing
during the intervening period. Alternatively, if the Enterprise is a
net investor in a given month, the marginal effect is reflected by its
after-tax earnings on new investments in Treasury bills.
This discounting procedure will reasonably relate changes in
capital to changes in an Enterprise's risk position. For example, if an
Enterprise were to take an incremental risk position that resulted in
an incremental loss during the first month of the stress period, that
loss would compound during the stress period at the Enterprise's after-
tax borrowing or investment rate. If an Enterprise is borrowing, this
one month's incremental additional loss would require additional
borrowings during the balance of the stress period. These additional
borrowings would create additional interest payments for which further
borrowing would be required. If the Enterprise is investing, the loss
would leave smaller amounts to be invested, which would earn less
interest. After applying the discount factors, the change in each
future month's capital would equal the initial loss. Thus, the change
in the estimated amount of the first month's incremental capital needed
to just maintain positive capital during the stress test would also
equal that initial loss. More generally, if a new asset were to
generate a stream of losses over the course of the stress period, the
amount of starting capital needed would rise by the present value of
this stream of losses.
IV. Technical Supplement
A. Purpose and Scope
This technical supplement provides detail on the specification and
estimation of statistical (econometric) models for mortgage
performance, and how those statistical models are applied in the
proposed risk-based-capital stress test. The supplement focus is on
technical aspects of the statistical modeling. This focus includes:
theoretical considerations, sources and uses of historical data,
functional forms for statistical models, development of explanatory
variables for the statistical analyses, results of statistical model
estimations, and application of the resulting statistical equations to
predict mortgage performance in the stress test. Each of the following
parts of this supplement covers these elements for its respective part
of mortgage performance. The topic areas covered here are:
Single Family Default/Prepayment,
Single Family Loss Severity,
Multifamily Default/Prepayment,
Multifamily Loss Severity, and
Property Valuation.
An additional, and important component of this Supplement is the
description of how the statistical models of mortgage performance are
reasonably related to the benchmark loss experience (BLE) identified in
NPR1. The first way in which OFHEO reasonably relates the mortgage
performance component of the stress test to the BLE is through
application of housing market conditions that represent the conditions
of that experience. Those conditions include house price growth rates,
rent growth rates, and rental vacancy rates. The next part of this
supplement, Property Valuation, details how OFHEO developed these
variables for use in the stress test. How these variables are actually
used in the stress test is covered in the section 3.5, Mortgage
Performance, of the Regulation Appendix, although some general
information is provided here.
The second way in which mortgage performance in general, and credit
losses in particular, are related to the BLE is through calibration
mechanisms that adjust statistically derived equations to match the
actual loss rates of the BLE. These adjustments are required because
the statistical equations are estimated over a wide range of data, of
which the benchmark experience is only a small part. To reasonably
relate mortgage losses to the BLE, the stress test imposes housing
market conditions from the time and place of the BLE. In addition, the
stress test adjusts defaults and severities by factors that cause the
test to replicate critical aspects of the BLE when the statistical
models are applied to benchmark loans. The methods of deriving these
calibration adjustment factors are described in the Single Family
Default/Prepayment and Single Family Loss Severity parts of this
Supplement.
B. Single Family Default/Prepayment
1. Introduction
To develop the stress test model of single family default and
prepayment rates, OFHEO analyzed the historical experience of
Enterprise single family loans from 1979 through 1995. This experience
is defined by an econometric model in which probabilities of default
and prepayment in each time period are
[[Page 18173]]
determined jointly using a multinomial logit specification. The
theoretical foundation used for choosing variables to use in the model
is financial options theory. This is the predominant theory used in
mortgage performance research. It suggests that borrowers make choices
regarding maintaining or terminating mortgages based upon the relative
financial value of those choices. In this context, each borrower has
the choice, in each time period, to make the payment and maintain the
mortgage, pay off the mortgage in full (a prepayment), or stop making
payments and default.
Owing to the large amount of data available to estimate this model,
OFHEO chose techniques that captured the essence of individual borrower
choice, consistent with efficient use of computer resources. These
techniques start with estimating separate sets of default and
prepayment equations for fixed-rate mortgages (FRMs) and for
adjustable-rate mortgages (ARMs).\196\ A third set of equations was
estimated to project the performance of less-prevalent single family
loan types relative to the dominant 30-year fixed-rate mortgages. The
second method of capturing borrower choice characteristics while
limiting computer resources was to use random samples of fixed-rate
loan products, rather than attempting to estimate the model on all
loans ever purchased by the Enterprises. The third method was to use
quarters rather than months as the observation time period. This time
period is important because each loan enters the analysis in the form
of an event history: every time period for which the loan was active
provides an observation for the statistical analysis. Using quarters
reduces the number of observations used in the statistical analysis
without losing any essential detail regarding borrower choices. The
last method of maintaining the quality of individual loan analysis
while limiting computer resources was to use a weighted regression
scheme, so that all loans do not need to enter the analysis
individually. All loans with the same characteristics are treated as
one loan, with the actual number of loans with those characteristics
used as a weighting factor.
---------------------------------------------------------------------------
\196\ In this model, ARMs include all mortgages that have
variable payment features.
---------------------------------------------------------------------------
The equations that result from the statistical analysis were
adjusted or calibrated to the BLE before use in the stress test. The
calibration procedure adjusts the default equations so that if the
actual benchmark loans (as defined in NPR1) were input into the
equations, with benchmark house price growth rates and interest rates,
the resulting 10-year cumulative default rate would identically match
that of the BLE (14.9 percent).
The remainder of this supplementary material is organized as
follows: Section 2 provides a summary of the conceptual framework
underlying the estimation of the statistical model of single family
mortgage default and prepayment. Section 3 describes the loan level
data used in the empirical analysis. Section 4 outlines the general
approach to the statistical analysis of default and prepayment events,
based on the application of the multinomial logit model. Section 5
defines the explanatory variables used in that analysis. The empirical
results are presented in section 6, which is followed in section 7 by a
discussion of the application of the estimated default and prepayment
equations in the stress test. Section 8 ends this supplementary
material by describing how the estimated model is used in the stress
test to produce results consistent with the BLE.
2. Conceptual Framework
Financial options theory is the most widely accepted theoretical
framework for the analysis of residential mortgage default and
prepayment. This framework hypothesizes that mortgage borrowers will
exercise embedded call (prepayment) or put (default) options when
either of these alternatives becomes financially optimal. The financial
options theory assumes that an individual mortgage borrower can
increase his lifetime wealth by defaulting on a mortgage when the
market value of the mortgage exceeds the market value of the house,
implying a direct empirical link between changes in housing values,
borrower equity, and the decision to default. Likewise, the option to
refinance the mortgage when market rates fall below the current rate on
the mortgage provides a means for borrowers to increase their wealth by
prepaying, and links observed prepayment behavior to changes in
interest rates.\197\
---------------------------------------------------------------------------
\197\ There may also be secondary effects of borrower equity on
prepayment, and of interest rates on default. For example, attempts
by borrowers to prepay their mortgages may be frustrated due to
declining house prices and failure to qualify for refinancing. On
the other hand, borrowers in a negative equity position may be
reluctant to default if they have current mortgage coupon rates that
are less than the prevailing market rate of interest. In this second
case, the asset value of the low interest rate mortgage would be
foregone if the put option is exercised and the borrower defaults.
However, the empirical significance of mortgage value for default is
questionable given the inability of borrowers to trade on this
asset, other than by selling the property and taking back a mortgage
at a rate between the original note rate and the current market
rate. This option is precluded by the ``due-on-sale'' provisions of
most residential mortgage contracts. The extent to which this option
is used informally is unknown.
---------------------------------------------------------------------------
Previous empirical studies on mortgage terminations have provided
empirical support for the options theory, as various approximations to
the financial values of the options have been found to be strongly
associated with observed default and prepayment outcomes.\198\ However,
some of the same studies also indicate that borrowers do not behave in
the ``ruthless'' manner suggested by the pure options theory. These
empirical studies vary in the degree to which the full implications of
the theory are incorporated, mainly due to limitations on the available
data and the ability to measure or impute options values to individual
borrowers.
---------------------------------------------------------------------------
\198\ Examples of empirical models based on the options
framework include: Dunn and McConnell (1981), Foster and Van Order
(1984, 1985), Buser and Hendershott (1984), Brennan and Schwartz
(1985), Kau, Keenan, Muller, and Epperson (1985, 1990), and
Hendershott and Van Order (1987).
---------------------------------------------------------------------------
The measurement of borrower equity has been addressed in
essentially two ways in the academic literature. One approach employs
stochastic simulations to impute aggregate distributions of properties
with positive or negative equity, while simultaneously accounting for
the impact of default and prepayment events on these distributions.
This is the approach used by Foster and Van Order (1984, 1985). Another
approach, adopted in recent work by Deng, Quigley, and Van Order (1996)
and Deng (1997), has been to combine mathematical assumptions about the
diffusion of housing values with loan-level data to assign ``ex ante''
probabilities of negative equity to individual properties.\199\ Both
approaches are generally consistent with the assumptions of the option
theory, and they differ mainly in their application to aggregate versus
loan-level data.
---------------------------------------------------------------------------
\199\ Probabilities assigned in this way are ``ex ante'' because
they depend only on information about individual mortgages available
at origination and subsequent changes in the mean (drift) and
variance (volatility) of house price appreciation rates. No
information on the incidence of default or prepayment among other
loans is used to adjust the projected distribution of housing values
used to assign probabilities of negative or positive equity to loans
that remain active.
---------------------------------------------------------------------------
In recent years, a consensus seems to have emerged among
practitioners that the option values, to the degree that they can be
measured, remain important for predicting default and prepayment,
[[Page 18174]]
but provide only necessary, rather than sufficient, conditions. For
example, in the case of mortgage default, negative equity alone may not
be sufficient to induce a borrower to default, but given some other
``trigger event,'' such as job loss or marital disruption, the decision
to default would then depend on whether equity was positive or
negative. In the case of prepayment, borrowers who would otherwise
appear to have a financial incentive to refinance (prepay) to obtain a
lower interest rate, may not wish to incur the associated transactions
costs given their expected time horizons for occupying the home.
While the option theory succeeds as a general framework, empirical
models of mortgage default and prepayment must be flexible enough to
account for variation in mortgage performance that may not appear to be
fully consistent with optimal behavior, such as borrowers defaulting
when house prices are increasing or prepaying when interest rates are
increasing. The empirical model must account for limitations on the
information available to compute the exact values of embedded options
for individual borrowers. In addition, a wide variety of loan
characteristics must also be accounted for, which has led to the
widespread application of what are generally referred to as ``options-
based'' empirical models, such as those cited above. The models applied
in the stress test are typical of those that use the options-based
approach.
3. Data
OFHEO obtained loan-level information on previous Enterprise single
family mortgage originations and used these data to estimate models of
mortgage performance. The data included information on the origination
characteristics of mortgages, information on last-paid installment
dates, and loan status outcomes from the Enterprise loan-tracking
systems. This information allowed OFHEO to reconstruct ``event
histories'' of the period-by-period performance of individual loans,
from the date of origination to either the point where the loan
terminated or the end of the sample period. OFHEO combined loan-level
information from both Enterprises to develop its own data files for
statistical analysis. Standardized or ``normalized'' data files were
constructed to assure similar content and structure across
Enterprises.\200\
---------------------------------------------------------------------------
\200\ The process of data normalization involved confirming the
consistency of mortgage product types and loan characteristics and
defining standardized data fields.
---------------------------------------------------------------------------
The options theory views mortgage default and prepayment events in
terms of decisions by individual borrowers to terminate their loans.
This view has implications for the way mortgage outcomes and their
associated probabilities are specified in the statistical analysis.
Default and prepayment are specified to occur in the month following
the date of the last-paid-installment. After mortgage prepayment, the
Enterprises are likely to update the loan status almost immediately. By
contrast, due to the varying length of the mortgage foreclosure
process, the Enterprises may not classify defaulting loans as defaults
until some months after the last-paid-installment date. However, in the
model, the default event is nevertheless considered to have occurred at
the point the borrower ceases payment on the loan.\201\ The event
history used for that loan ends at that point in time. The data used in
the statistical analysis included mortgage originations for the period
from January 1979 to December 1993, with mortgage performance measured
through December 1995. Therefore, these data provided a minimum of two
years of loan experience for the most recent origination cohorts.\202\
---------------------------------------------------------------------------
\201\ At the time that data bases were constructed for this
analysis, information was not available from Freddie Mac on last-
paid-installment dates. Therefore, OFHEO used the ``closing date''
for Freddie Mac's defaulted loans. This is the date of disposition
of a foreclosed property. The last-paid-installment date was used
for Fannie Mae defaults.
\202\ Note that for some loans the last-paid-installment will
occur prior to the end of the sample, with no corresponding change
in loan status from active to defaulted. These ``censored'' events
were treated in the same manner as loans that remained active
through the end of the sample period. That is, they are viewed as
active up to and including the last quarter in the sample period.
Note that these censored default events do not occur in sufficient
numbers to have a material impact on the statistical estimates. One
reason is that during those time periods and places in which the
incidence of default was greatest, such as, for example, in the
historical benchmark experience, foreclosure and changes in loan
status occurred within several months of the last payment by the
borrower. In addition, relatively complete loan histories are
available for those loan origination cohorts among which the
majority of default events occurred on Enterprise loans. While more
recent cohorts with shorter event histories have greater potential
for censoring of default events, the impact of censoring on the
statistical estimates is negligible because default rates have been
so low in recent years.
---------------------------------------------------------------------------
Ideally, models would be estimated using contemporaneous values of
factors predictive of default and prepayment during each period a loan
is outstanding. Although this type of ``panel'' data does not exist for
historical Enterprise loan records, it was possible to reconstruct
historical data on key determinants of default and prepayment, such as
house prices and interest rates, and add this information to the
individual loan event histories. Using these histories, OFHEO was able
to estimate dynamic models for default and prepayment. The models are
``dynamic'' in the sense that OFHEO can estimate and simulate mortgage
performance in response to actual or hypothetical (e.g., stress test)
changes in economic circumstances over time.
4. Specification of the Statistical Model
The proposed regulation employs a monthly cash flow model of
Enterprise performance over a ten-year stress period. The simulation of
mortgage cash flows requires conditional rates of default and
prepayment to be applied to outstanding mortgage balances during each
month of the stress test. The purpose of the models described in this
technical supplement is to provide a means of generating the required
termination rates in a manner that is reasonable for Enterprise loans
under the circumstances of the stress period.
Conditional rates of default and prepayment vary depending on a
variety of factors, both random and systematic, some of which are fixed
at origination and others that vary over time. Characteristics of loans
and borrowers at origination can affect the level and timing of
mortgage default and prepayment throughout the life of the loan. For
example, conditional default and prepayment rates exhibit
characteristic age-profiles that increase during the first years
following origination, peak sometime between the fourth and seventh
years, and decline gradually over the remaining years.\203\ Default and
prepayment rates also vary systematically in response to economic
circumstances and other factors over time, such as changes in house
prices and interest rates that affect the value to the borrower of
embedded options.
---------------------------------------------------------------------------
\203\ See discussion in Schwartz and Torous, at 379 (1989).
---------------------------------------------------------------------------
Like other time-or age-dependent processes, mortgage terminations
are highly amenable to analysis using statistical survival-time models
specified in terms of conditional probabilities of prepayment and
default. Default and prepayment are ``competing risks,'' which means
that the occurrence of one type of event precludes the chance to
observe when the other event might have occurred, and vice versa. In
such a case it is necessary to account for the joint mathematical and
statistical dependence of the conditional probabilities of default and
prepayment on each other. Failure to account for the competing-risks
nature of the events can lead to projections of total termination
[[Page 18175]]
rates (default plus prepayment) that are mathematically inconsistent
and that would preclude their application in the type of actuarial
calculations of cash flows required for the stress test.
As outlined above, mortgage default and prepayment result in an
observed last-paid-installment, after which no further payments are
forthcoming. Thus, for loans outstanding at the beginning of each time
period, three mutually exclusive outcomes are possible in the model:
(1) the borrower defaults; (2) the borrower prepays the loan in full;
or (3) the borrower makes the scheduled loan payment, and the loan
remains active and part of the event history sample for the next time
period. For the purposes of the statistical analysis, each of these
outcomes is interpreted as an ``event.'' This approach implies that
each loan contributes potentially many observations to the event
history sample, depending on how long it remains active before
experiencing one of the terminal events or reaching the end of the
sample period.
a. Multinomial Logit Models
OFHEO has estimated multinomial logit models for quarterly
conditional probabilities of default and prepayment.\204\ Several
empirical studies have applied some form of the logit or similar
qualitative response models to analyze mortgage prepayment and default
behavior.\205\ The corresponding mathematical expressions for the
conditional probabilities of default (D(t)),
prepayment (p(t)), or remaining active
(A(t)) over the time interval from t to t + 1 are
given by:
[GRAPHIC] [TIFF OMITTED] TP13AP99.001
[GRAPHIC] [TIFF OMITTED] TP13AP99.002
[GRAPHIC] [TIFF OMITTED] TP13AP99.003
Constant terms D and p, and
coefficient vectors D and p,
are the unknown parameters that must be estimated. XD(t) is
a vector of mostly time dependent explanatory variables that are
assumed to influence directly the conditional probability of defaulting
(versus remaining active), and Xp(t) is a vector of mostly
time dependent explanatory variables assumed to influence directly the
conditional probability of prepaying (versus remaining active).\206\
The probability of remaining active (A(t)) is equal
to 1 minus the other two probabilities, so that the three probabilities
sum to 1.
The probabilities and coefficient vectors have a convenient
interpretation when expressed in terms of odds ratios:
[GRAPHIC] [TIFF OMITTED] TP13AP99.004
[GRAPHIC] [TIFF OMITTED] TP13AP99.005
These expressions imply that the percentage impact of a one-unit
change in an element of XD(t) on the relative probability or
odds of defaulting versus remaining active is given by the
corresponding element of the coefficient vector, D.
A similar result holds for prepayment. Note also, that while changes in
variables that affect the probability of prepayment affect the absolute
level of the probability of default, and vice versa, such changes
affect the probability of remaining active in a symmetric manner, so
that the ``odds'' of defaulting versus remaining active are not
affected.\207\
---------------------------------------------------------------------------
\204\ The decision to model default and prepayment as quarterly
events was consistent with the application of quarterly house price
indexes in computing the underlying distributions of borrower
equity. The resulting quarterly default and prepayment probabilities
were converted to monthly factors for input to the monthly cash flow
calculations required for application in the stress test.
\205\ Examples of previous applications of the logit model are
Campbell and Dietrich (1983), Zorn and Lea (1989), and Cunningham
and Capone (1990).
\206\ Some elements of XD(t) and Xp(t) are
constant over the life of the loan and are not functions of t.
\207\ The multinomial logit model is widely applied in the
analysis of consumer choice among discrete alternatives, where this
feature has been called the ``independence of irrelevant
alternatives.'' In the context of consumer choice theory this
independence can result in apparent anomalies when close substitutes
to existing choices are introduced. See, for example, McFadden
(1976). This issue does not arise in the present context.
---------------------------------------------------------------------------
[[Page 18176]]
b. Estimation of Multinomial Logit Coefficients
The multinomial logit specification given by equations (1)-(3) is a
purely mathematical representation of the underlying probabilities. How
the unknown parameter coefficients of the logit model are estimated
statistically depends on whether the model is applied to individual or
aggregate data. Under some circumstances, the two approaches are
mathematically equivalent. However, in some situations, the use of
aggregate data may entail considerable loss of information.\208\
---------------------------------------------------------------------------
\208\ For example, if the data are aggregated by taking average
values of the explanatory variables within broad product groupings,
then particular combinations of explanatory variables that exist for
individual loans and which are associated with significant
differences in probabilities of default and prepayment, will not be
represented in the data. While this may not matter under ``normal''
circumstances, it could limit the usefulness of the model in
projecting rates of default and prepayment within high risk
categories under circumstances different than those embodied in the
original aggregation scheme, such as those of the stress test.
---------------------------------------------------------------------------
If only aggregate data were used, the proportions of loans
defaulting, prepaying, and remaining active would be used to estimate
the unknown coefficients D, p,
D, and p directly by replacing
the probabilities in equations (4) and (5) with the corresponding
observed sample proportions and applying ordinary least squares. In
this case the explanatory variables XD(t) and
Xp(t) correspond to the characteristics of the groups or
classes of loans used in tabulating the observed sample proportions.
When loan-level data are available, it is possible to use equations
(1)-(3) as an exact mathematical representation of the probabilities of
individual loan events. In this case, estimation of unknown
coefficients is achieved by the method of maximum likelihood. This
approach chooses the values of D,
D, p, and p
that maximize the joint likelihood or probability of the entire event-
history sample having actually occurred. For example, the joint sample
likelihood is the product of the probabilities of each of the
independent loan event observations:
[GRAPHIC] [TIFF OMITTED] TP13AP99.006
where for each observation i = 1,2. . ., N, Pt is the
estimated probability that the event that is actually observed would
have occurred. These probabilities are obtained by substituting the
appropriate expression from equations (1)-(3) for Pi in
equation (6). The solution is found by varying the values of the
elements of D, D,
p, and p until the joint
probability reaches its maximum value. The final values of
D, D, p,
and p are the maximum likelihood estimates.
Numerous statistical software packages exist for this purpose.
The approach adopted by OFHEO is based on loan-level data, which
has the significant advantage of preserving as much detail as possible
on individual loan circumstances. This approach results in a flexible
description of loan behavior, which can be used to project mortgage
performance under the abnormal scenarios of the proposed regulation.
5. Explanatory Variables for Default and Prepayment
OFHEO estimated three separate sets of multinomial logit
probability equations. The primary default and prepayment equations are
for single family, 30-year FRMs. These loans comprise about 80 percent
of all single family loans in the historical data obtained from the
Enterprises. A second set of equations was estimated solely on data for
ARMs. All loan types with any potential payment adjustments throughout
the life of the loan were included as ARMs for purposes of the
statistical estimation. A third set of default and prepayment equations
was estimated to project the performance of less-prevalent single
family loan types relative to 30-year fixed-rate mortgages. This
estimation was performed using data on 30-year FRMs and all other
fixed-rate loan types (including balloons). These loan types were
grouped as: 20-year FRM, 15-year FRM, balloon, FHA/VA, and second
liens. Data on 30-year FRMs are included in the estimation sample
because the number of observations on other, less popular fixed-rate
mortgage types was insufficient for estimating product-specific default
and prepayment equations. However, the resulting default and prepayment
equations are only used to project performance of the alternative
product types, and not 30-year FRMs.
All three statistical estimations use the same conceptual
underpinnings and empirical specifications, and only vary based on the
data samples used in estimation. Thus, the basic definitions of the
variables are the same across all three sets of equations, although the
way some of the interest rate variable values change over time will
differ, for example, for FRM loans and ARM loans, because of
differences in their contractual terms.
For convenience, we refer to the three separate data sets and
statistical estimations as model 1 (30-year FRMs), model 2 (ARMs), and
model 3 (all fixed-rate products). In addition to the basic set of
explanatory variables included in all three models, model 3 includes
product-specific adjustment constants. The adjustment constants act
like multipliers to the baseline default (hazard) rates of 30-year
FRMs. The impacts of all other explanatory variables are presumed
constant across product type, so there are no product-type adjustments
to their coefficients. Because ARMs are believed to perform differently
than FRMs, due to changing payments over time, they are treated in a
separate estimation (model 2) so that variable coefficients can be
uniquely identified for ARM versus FRM loans.
The explanatory variables XD(t) and Xp(t)
used to estimate the unknown coefficients of the multinomial logit
models are listed in Table 31. All of the variables except mortgage age
(AGE) were coded as categorical variables. Categorical variables are
advantageous for several reasons. For instance, assigning the various
explanatory variable outcomes to categories allows one to estimate
effects that may be non-linear without having to experiment with many
different functional forms. Because each categorical explanatory
variable has minimum and maximum categories (determined through
observation of the historical data), the
[[Page 18177]]
impact of particular variables on rates of default or prepayment
projected from the model is constrained to be within previous
historical experience.\209\ This helps to avoid unreasonable
extrapolations when projecting mortgage performance under stress test
conditions. Another advantage of using categorical outcomes for the
explanatory variables is that it anticipates the need to apply the
models to aggregated loan groups in the stress test.\210\ The benefit
of starting with loan-level data is that it allowed OFHEO to develop
both the explanatory variables and stress test loan groups in a
consistent manner, thus minimizing the loss of information due to data
aggregation.
---------------------------------------------------------------------------
\209\ This constraint applies specifically to the marginal
contribution of particular explanatory variable outcomes, not to the
overall level of the default and prepayment probabilities projected
by the model. For example, if several explanatory variables
simultaneously take on values that have not been previously observed
in combination, then it is possible that the projected probabilities
of default or prepayment would exceed those observed in the
historical data. This type of outcome is anticipated by the 1992
Act, which requires regional adverse credit conditions to apply
nationally to all loans at the same time.
\210\ The loan groups used in the stress test were developed in
conjunction with the classification of explanatory variable outcomes
in the statistical analysis of mortgage default and prepayment.
Aggregation of mortgage assets in the stress test recognizes the
need to classify assets within broad product categories for
financial accounting. Within the context of the proposed regulation,
the use of aggregate loan groupings also facilitates the assignment
of new loan products to existing categories with known risk
characteristics. Further explanation of the aggregate loan groups
used in the stress test is in section III. A., Mortgage Performance
of the preamble.
---------------------------------------------------------------------------
The summary of explanatory variables starts with descriptions of
the two key options-related predictors of mortgage default and
prepayment-respectively, the probability of negative borrower equity
and the mortgage premium value. A review of additional interest rate
variables and loan characteristics that are used as explanatory
variables follows.
a. Probability of Negative Equity
The put option has value to the borrower when the property is worth
less than the outstanding balance on the mortgage. In that case, the
borrower is in a negative equity position. Thus, the equity position of
the borrower is determined by the difference between the market value
of the property securing the loan, P(t), and the unpaid mortgage
balance, UPB(t):
[GRAPHIC] [TIFF OMITTED] TP13AP99.007
Ideally, periodic observations on the values of individual
properties would be used to update individual house values and borrower
equity at the same frequency (monthly) at which the decision to prepay
or default can be exercised. However, because individual housing values
are not updated continuously it is not possible to compute updated
values of EQ(t) for individual borrowers with sufficient accuracy for
this measure to be used directly at the loan level.\211\
---------------------------------------------------------------------------
\211\ As discussed above, given the measurement difficulties
associated with borrower equity at the loan level, some researchers
have used various means of simulating the distribution of borrower
equity. For example, Foster and Van Order (1984, 1985) used a Monte
Carlo simulation of a synthetic mortgage pool in conjunction with a
house price diffusion process and actual default and prepayment
rates to reconstruct a time-series for the number of borrowers in a
negative equity position. Under additional restrictions on the model
(i.e., that only borrowers with negative equity default, and only
borrowers with positive equity prepay), the time-series for the
number of borrowers with negative equity (various levels) was used
in regressions for conditional default and prepayment probabilities.
---------------------------------------------------------------------------
It remains possible, however, to characterize the equity positions
of individual borrowers in terms of ex ante probabilities of negative
equity.\212\ The probability of negative equity is a function of the
scheduled current loan balance and the likelihood of individual house
price outcomes that lie below this value. Projected distributions of
individual housing values relative to the value at mortgage origination
were calculated by applying estimates of house price drift and
volatility obtained from independent estimates based on the OFHEO House
Price Index (HPI).\213\
---------------------------------------------------------------------------
\212\ See the discussion of ex ante probabilities of negative
equity in footnote 199.
\213\ House price drift is defined here as the average rate of
house price appreciation as determined by the appropriate market
house price index, while volatility is defined as the variance in
individual house price appreciation rates around the market average
rate of appreciation.
---------------------------------------------------------------------------
The required estimates of house price drift and volatility are
direct by-products of the estimation of the OFHEO HPI. The OFHEO HPI is
based on a modified version of the weighted-repeat-sales (WRS)
methodology (Case and Shiller, 1987, 1989), and is consistent with the
assumption that housing values are generated by a log-normal diffusion
process. This means that over time individual housing values will
appreciate at different rates, distributed randomly around the average
rate of appreciation. Over time, the cumulative rates of appreciation
for individual homes will become more and more dispersed or diffused,
hence the reference to diffusion processes. Mathematically, individual
house prices are assumed to obey a non-stationary log-normal diffusion
process in which individual house price appreciation since mortgage
origination is normally distributed with variance 2
(A) around the expected rate of appreciation from the HPI,
(t), computed as:
[GRAPHIC] [TIFF OMITTED] TP13AP99.008
Where A is loan age (in quarters), and HPI(0) is the value of the
HPI at time of loan origination.\214\ For the individual borrower with
original house price P(0) at time 0, the probability of negative equity
at time t, PNEQ(t) is given by:
---------------------------------------------------------------------------
\214\ Estimates of expected appreciation or drift in house
prices are obtained directly from the estimated values of the HPI
for each of the nine U.S. Census divisions. Estimates of diffusion
volatility, 2(A), are computed using the
estimated parameters for the error variance of individual log-
differences in housing prices that are obtained from the second-
stage of the WRS method for each division. See Calhoun (1996) for
additional details. Deng, Quigley, and Van Order (1996) applied a
similar approach using WRS indexes for 26 metropolitan areas
estimated using Freddie Mac data.
[[Page 18178]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.009
[GRAPHIC] [TIFF OMITTED] TP13AP99.010
where (x) is the standard normal cumulative distribution
function evaluated at x. This expression quantifies the relationship
between changes in house prices on average, and the likelihood of
negative appreciation on individual properties that places some
fraction of borrowers in a negative equity position. The imputed share
of borrowers with negative equity implied by equation 10 is used as a
proxy for the probability of negative equity for an individual
borrower.\215\ The computed probabilities of negative equity are
assigned to one of eight categorical outcomes, as summarized in Table
31.
---------------------------------------------------------------------------
\215\ Although the market level (regional) values of house price
drift and volatility are used, the imputed probability of negative
equity is still specific to the individual borrower's circumstances,
since the loan-specific values of original LTV and loan amount are
used in the calculations.
---------------------------------------------------------------------------
b. Relative Spread
The theoretical value of the call (prepayment) option on a mortgage
is a function of the difference between the present value of the future
stream of mortgage payments discounted at the current market rate of
interest, R(t), and the present value of the mortgage evaluated at the
current note rate, C(t). The actual value of this call option to the
borrower is unknown due to uncertainty over the future time path of
mortgage payments associated with uncertain future probabilities of
prepayment and default. Therefore, it is common to use other variables
to capture the impact of the call option value on prepayment rates.
Following recent work by Deng, Quigley and Van Order (1996), OFHEO
approximated the call option value using the relative spread variable,
RS(t):
[GRAPHIC] [TIFF OMITTED] TP13AP99.011
Positive values of the call option exist when the mortgage coupon
exceeds the current market interest rate (positive spread), and the
borrower can benefit financially by refinancing to obtain a lower
interest rate. Outcomes for the relative spread variable are classified
into seven categorical outcomes, as summarized in Table 31.
c. Prepayment Burnout
Recent studies of mortgage terminations have emphasized the
importance of previous interest rate environments for distinguishing
among borrowers more or less likely to exercise the prepayment option
when the opportunity arises.\216\ The tendency for the most responsive
borrowers to prepay first, so that the remaining sample of borrowers
are those with lower average conditional probabilities of prepayment,
contributes to the observed seasoning or ``burnout'' of mortgage pools.
The indicator variable B(t) is included to measure whether the borrower
has missed a previous refinance opportunity.\217\ B(t) is defined by
whether the market rate of interest was 200 basis points or more below
the coupon rate of the mortgage during two or more quarters over the
past two years. Those who have missed previous refinance opportunities
are predicted to have lower conditional probabilities of prepayment and
higher conditional probabilities of default. Failing to refinance under
favorable interest rate conditions may indicate the existence of other
credit-related problems, such as failure to obtain an adequate property
appraisal.\218\
---------------------------------------------------------------------------
\216\ For example, see the discussions of borrower heterogeneity
and path dependence in Bartholomew, Berk, and Roll (1988), and the
discussion of burnout in Richard and Roll (1989).
\217\ The indicator variable equals one if the spread between
the note rate on the mortgage and the quarterly average market rate
of interest has been 200 basis points or greater during any two of
the past eight quarters.
\218\ See footnote 198.
---------------------------------------------------------------------------
d. Yield Curve Slope
Expectations about future interest rates and differences in short-
term and long-term borrowing rates associated with the slope of the
Treasury yield curve influence the choice between ARM and FRM loans and
the timing of refinancings and prepayments. A high value for the slope
of the yield curve indicates relatively favorable short-term rates,
increasing the likelihood that a borrower refinances to an ARM to take
advantage of the lower initial coupons that can be offered by lenders.
The variable YS(t) is included to measure the current slope of the
yield curve. This variable is computed as the ratio of the ten-year
Constant Maturity Treasury yield (CMT) to the one-year CMT, and
assigned to four categorical outcomes.
e. Mortgage Age
The existence of other demographic and economic processes that may
``trigger'' mortgage default or prepayment, and the inability to
measure the diffusion of house prices and the distribution of borrower
equity precisely, create a need to account directly for age-specific
differences in conditional rates of default and prepayment.\219\ The
direct dependence of the conditional probabilities on mortgage age
recognizes the existence of other borrower processes and unobserved
heterogeneity that induce duration dependence in the conditional rates
of termination and help to explain the typical age patterns of default
and prepayment.\220\ For this reason,
[[Page 18179]]
mortgage age (AGE) is included as an additional explanatory variable in
the empirical model. The model utilizes a quadratic function of
mortgage age, where age is defined as the number of quarters since
origination. The use of a parametric function of age instead of
categorical values is based on two considerations. First, the use of
categorical age values for individual quarters would result in a large
number of additional coefficients to estimate. Combining loans into
broader age groupings to reduce the number of parameters can produce
large differences in rates of default and prepayment with small
increments in age for loans graduating from one age category to the
next. Second, when individual age categories are used, they show that a
quadratic age function is a reasonable assumption, at least for the
first eight to ten years. At higher values of mortgage age, the samples
are much smaller (most loans have terminated by these ages), with the
result that the estimates for individual age categories are quite
erratic due to sampling error. The use of a simple functional form like
the quadratic helps to smooth the estimates of the age effects for the
higher age groups.
---------------------------------------------------------------------------
\219\ Under a pure options model, the typical age patterns of
conditional default and prepayment rates might be attributed
entirely to the diffusion of housing values and the introduction of
unobserved differences (heterogeneity) in the equity positions of
individual borrowers, resulting in differences in the rates of
default and prepayment among particular subsets of individual
borrowers. As these differences emerge following mortgage
origination, the observed average conditional default and prepayment
rates will initially increase. Eventually, as ``high risk''
borrowers depart the sample or mortgage pool, the average
conditional rates of default and prepayment will decline.
\220\ See Lancaster (1990) for a discussion of the impact of
unobserved heterogeneity on estimates of duration dependence in
econometric models of transition probabilities. Other borrower
processes include residential mobility, employment mobility,
involuntary unemployment, and demographic events related to
household formation and dissolution, mortality, and fertility.
Ideally, given suitable household-level data, these other processes
would be modeled jointly with mortgage terminations.
---------------------------------------------------------------------------
f. Original LTV
The original LTV ratio, LTV(0), serves as an indicator of the
income and net worth of the borrower at mortgage origination, and
directly determines the initial equity position of the borrower. To the
extent that income and wealth are negatively correlated with LTV(0),
high LTV borrowers will have fewer economic resources to finance the
transactions costs of prepayment or endure spells of unemployment or
other trigger events that might otherwise cause them to exercise the
default option in a sub-optimal manner. Finally, high LTV borrowers
have already demonstrated a willingness to ``leverage'' the financing
of the home purchase, which may portend a greater sophistication or
``ruthlessness'' in the exercise of the default option. Thus, one would
expect higher rates of default and lower rates of prepayment as LTV(0)
increases. The six LTV(0) categories used in the default/prepayment
models are similar to those used by the Enterprises in their annual
reports and information statements.
g. Season of the Year
The variable SEASON(t) was included to account for the current
season (quarter) of the calendar year, in recognition of the potential
impact of weather, school schedules, and seasonal employment patterns
on residential mobility and default and prepayment probabilities.
h. Occupancy Status
OS is an indicator variable included to distinguish mortgages on
owner-occupied units from investor loans. Owner occupants should be
less likely than investors to exercise the default option given the
direct benefits they receive from the consumption of housing services.
Owner occupants should be more likely to prepay than investors for non-
financial reasons such as residential mobility.
i. Relative Loan Size
The ability to bear the transactions costs of refinancing, or to
weather economic stress and avoid default, will be correlated with the
income level of the household. Given the lack of information in the
historical data on household income at origination, a measure of
relative loan size provided a proxy for the relative income level of
the household. LOANSIZE was defined as the ratio of the original loan
amount relative to the average-sized Enterprise loan originated in the
same State during the same origination year.\221\
---------------------------------------------------------------------------
\221\ Price Waterhouse (1990) reported significant differences
in claim rates for FHA mortgages stratified by loan size. Smaller
loans were observed to fail at significantly higher rates than other
loans.
---------------------------------------------------------------------------
j. Product Type Indicators
Five product type indicators were created to account for the
performance of non-standard loans relative to the standard 30-year FRM
loans in model 3: 20-Year FRM, 15-Year FRM, balloon, FHA/VA, and
seconds. These indicator variables provide the adjustment constants
mentioned earlier.
k. ARM Coupon Rate Dynamics
To estimate the current values of both the probability of negative
equity, PNEQ(t), and the relative spread, RS(t), variables for ARM
loans, it was necessary to trace the path of current coupon rates over
the active life of individual mortgages. For standard ARM products, the
coupon rate resets periodically to a new level that depends on the
underlying index, plus a fixed margin, subject to periodic and lifetime
interest rate caps that specify the maximum and minimum amounts by
which the coupon can change on any one adjustment and over the life of
the loan.\222\ ARM coupon rates are updated using the following
formula:
---------------------------------------------------------------------------
\222\ Detail on specific ARM contracts was obtained in some
cases from loan-level information, and in other cases was obtained
using plan-level detail for loans in certain ARM product categories.
Any loan product with variable interest rates was classified as an
ARM, and modeled according to product terms. This includes so-called
two-step mortgages and mortgages with interest-rate buydowns. For
simplicity, the margin was set at 2 percent for all ARMS.
[GRAPHIC] [TIFF OMITTED] TP13AP99.012
Where Index (t) is the underlying index value at time t, S is the
``lookback'' period, and Margin is the amount added to Index (t--S) to
obtain the ``fully-indexed'' coupon rate. The periodic adjustment caps
are given by PeriodUpCap and PeriodDownCap, and are multiplied by an
indicator variable A(t) which equals zero except during scheduled
adjustment periods. The maximum lifetime adjustments are determined by
and LifeUpCap and LifeDownCap.\223\
---------------------------------------------------------------------------
\223\ The majority of Enterprise ARM loans are indexed to the
one-year Treasury rate, with smaller but significant numbers indexed
to either the five-year or ten-year Treasury rate, the 11-District
Cost of Funds Index (COFI), or the London Inter-Bank Offer Rate
(LIBOR). A small percentage of ARM loans are indexed to the six-
month or three-year Treasury rates. The majority of ARM loans had
lifetime adjustment caps of five or six percent, and have no
lifetime rate floors. Most have periodic rate adjustment caps of two
percent, while some have periodic rate adjustment caps of one
percent. The majority of ARM loans have adjustment frequencies of
one year, while a significant minority are adjusted every six
months.
---------------------------------------------------------------------------
[[Page 18180]]
6. Empirical Results
The three models were estimated by the method of maximum likelihood
using the SAS CATMOD procedure. The CATMOD procedure
employs a design matrix that automatically converts all categorical
variables to a series of indicator variables prior to estimation. As
discussed above, all explanatory variables except mortgage age were
converted to indicator variables. This allows one to reduce the data to
a smaller number of loan records, each representing unique combinations
of the categorical variables, to which a frequency count is assigned
and applied as a sampling weight in subsequent statistical analyses.
This approach avoids the need to undertake choice-based sampling (e.g.,
over-sampling of defaulted loans) in order to assure that sufficient
numbers of rare events like mortgage default are obtained.\224\
However, given the large number of loan level observations available to
OFHEO, simple random samples were used to estimate the 30-Year FRM and
Multiple Products models. All available data were used to estimate the
ARM model.\225\
---------------------------------------------------------------------------
\224\ It has been demonstrated for static logit models that
choice-based sampling results in biased estimates of the
coefficients of the logit constant terms, for which relatively
simple corrections are available, based on the population
distribution of the explanatory variables across groups defined by
dependent variable outcomes (Costlett, 1981). It is not clear that
the same form of correction applies to the retrospective event-
history sample used in this analysis. Selection on the basis of
default outcomes implies selection of an array of preceding ``non-
events'' for each quarter the loan was active, so that the
distributions of the explanatory variables for specific age
categories depends on the timing of default events for individual
loans.
\225\ A ten-percent random sample was used for the 30-Year FRM
model and the Multiple Products model. All data used for estimation
were subject to a variety of data quality screens and available data
for all the explanatory variables.
---------------------------------------------------------------------------
Table 32 contains the parameter estimates for the three
models.\226\ The constant and age parameters are listed first, as they
provide a baseline function to which the effects of other variables can
be added. There is a high level of consistency in the coefficient
estimates across all three models, and all three models provide
empirical support for the importance of the options-related variables.
---------------------------------------------------------------------------
\226\ Note that a particular feature of the SAS CATMOD procedure
is that when it estimates the coefficients corresponding to a
variable with N categories, the program estimates only the first N-1
coefficients. The final-category coefficient for each variable is
computed as the additive inverse of the first N-1 category
coefficients.
---------------------------------------------------------------------------
The coefficient estimates for the probability of negative equity
variable (PNEQ) vary on the same order of magnitude for default as the
coefficient estimates for the original LTV variable. PNEQ is also
important for prepayment, in the opposite direction, consistent with
the expectation that those most likely to have negative equity will
have the greatest difficulty selling their homes or refinancing their
mortgages, and therefore be less likely to prepay their existing
mortgages. Original LTV is relatively unimportant for prepayment,
although those in the lowest LTV category are more likely to prepay.
The value of the call option measured by the relative spread (RS)
shows quite large effects on prepayment in the hypothesized direction.
The higher the coupon rate on the mortgage relative to the current
market rate of interest the higher the likelihood of prepayment. Note
the general similarities between the RS coefficient estimates for
models one and two (30-year FRMs and ARMs). Because ARM coupon rates
will adjust with changes in market rates, ARM borrowers are less likely
than FRM borrowers to end up with large positive or negative RS values.
However, the estimates in Table 32 imply that ARM and FRM borrowers
behave in a similar manner under comparable values of the call option.
The prepayment burnout variable, B, is most important for default
rates, and indicates that missed opportunities to prepay are associated
with higher credit risk. This result reinforces the results discussed
above for PNEQ, where higher values of PNEQ were associated with lower
probabilities of prepayment. This result also reflects the lack of
precision in measurements of borrower equity at the loan level.
The slope of the yield curve (YS) is important for the probability
of prepayment for FRM borrowers, especially for steep positive values
of the slope. This result is consistent with the tendency of borrowers
to refinance to ARM mortgages when short-term rates are relatively low
and lenders can offer very favorable initial coupons (``teaser''
rates). It is also consistent with the assumption that the expectation
of higher interest rates in the future may cause some borrowers to
refinance sooner to lock in lower rates. The yield curve slope variable
has similar, but smaller, effects for ARM borrowers.
The SEASON variable has modest effects in the anticipated
directions. For FRM borrowers, prepayment rates are lower than average
in the Winter and higher in the Spring. Default rates are lower in the
Winter and higher in the Fall. For ARMs, prepayments are also higher in
the Fall, but defaults are lower in that season.
Occupancy status (OS) has much larger impacts on default
probabilities for ARM borrowers than FRM borrowers. For both product
types, investors are more likely to default than owner-occupants, and
much more so for ARM borrowers than FRM borrowers. It is reasonable to
expect that owner-occupants will be less ruthless in the exercise of
the default option given the offsetting value they receive from living
in the home. The prepayment effects are more similar across ARM and FRM
borrowers.
The variable LOANSIZE was included as a proxy for borrower income
at origination. The results in Table 32 indicate that relative loan
size is not particularly important for default probabilities, at least
after controlling for the other explanatory variables. LOANSIZE is much
more important for prepayment, with smaller loans prepaying at lower
rates than relatively large loans. This is consistent with the
interpretation of LOANSIZE as a proxy for borrower income. Lower income
borrowers may lack the resources to bear the transactions costs of
refinancing, causing them to prepay at lower rates than higher income
borrowers with relatively large loans. Lower income borrowers may also
be less mobile than higher income borrowers. The results for prepayment
are similar across FRM and ARM borrowers.
The results for the two fixed-rate models, models one and three,
are generally quite consistent. The individual product type indicators
in model 3 provide estimates of the relative rates of default and
prepayment of various fixed-rate products in comparison to 30-Year
FRMs, and in comparison to each other. Balloon mortgages have the
highest rates of default and prepayment relative to 30-Year FRMs.
Intermediate FRM products (15-Year and 20-Year) default at lower rates
than 30-Year FRMs. This result is consistent with more rapid loan
payoff and accumulation of borrower equity for these borrowers. Rates
of prepayment on intermediate FRMs are comparable to those on 30-Year
FRMs. FHA and VA loans have higher rates of default and lower rates of
prepayment than 30-Year FRM loans. Results for the category of second
loans is most similar to the FHA/VA loans.
7. Application of the Models in the Stress Test
The three product-based single family models provide the means to
project the conditional default and prepayment probabilities required
as inputs to the cash flow model of Enterprise financial
[[Page 18181]]
performance. The stress test aggregates single family loan-level data
into loan groups based on the following characteristics: Enterprise,
portfolio (securitized vs. retained), product type, origination year,
original LTV ratio class, original coupon class, starting coupon class,
and region (Census division). The information contained in
characteristics data for each aggregated loan grouping is sufficient,
when combined with data on house price growth rates and interest rates,
to compute and update all of the explanatory variables needed for
computing conditional default and prepayment probabilities during the
stress period.
There are three exceptions to this general statement. The variables
SEASON and LOANSIZE were not used to classify loans for the purpose of
the stress test. The SEASON variable was excluded when applying the
logit models to project default and prepayment probabilities over the
stress period.\227\ The LOANSIZE variable was retained, but all loans
were categorized as being of average size. These two changes reduced by
a factor of nine the number of loan groups that had to be processed
when running the stress test. Accounting for seasonal effects and
differences in default and prepayment rates by loan size was not
considered essential for projecting mortgage performance in the stress
test.\228\ In addition, the variable OCCUPANCY, used to distinguish
mortgages on owner-occupied units from investor loans, is replaced by
the portfolio average percentages for each occupancy status. Thus,
instead of creating separate loan groups for owner-occupied and
investor loans, these loans are combined into a single group, and a
weighted average of the logit coefficients for owners and investors is
used when projecting default and prepayment probabilities. This
procedure reduces the number of records that must be processed by a
factor of 2, but still allows OFHEO to account for changes over time in
the percentage of Enterprise mortgages that are investor loans.
---------------------------------------------------------------------------
\227\ The parameter estimates generated by the SAS CATMOD
procedure are defined so that they sum to zero across all categories
of a given explanatory variable. This implies that dropping them
from the model is equivalent to assuming that the logit
probabilities for default and prepayment include the average effect
across all the possible categories of the excluded variable.
\228\ Including the SEASON variable in estimation can be
justified because it helps to isolate the statistical impact of
changes in house prices on borrower equity from purely seasonal
fluctuations in default and prepayment rates. Likewise, LOANSIZE and
original LTV are both likely to be related to borrower income and
wealth at mortgage origination. However, because LOANSIZE is defined
relative to the average sized loan within a state in the year of
origination it provides a somewhat different measure of relative
income or wealth.
---------------------------------------------------------------------------
The detail contained in the starting position loan group records is
sufficient to treat each loan group as if it performs like a single
loan, with the projected probability of default or prepayment from the
model corresponding to the share of the loan group balance that will
default or prepay in any given period (i.e., by the ``law-of-large-
numbers''). Group-specific average values of original LTV and mortgage
coupon are used in place of exact loan-specific values in computing
explanatory variables requiring these as inputs (e.g., PNEQ and RS).
Categorical values such as original LTV and region (Census division)
are classified in the same way for both the loan-level data used for
estimation and the loan groupings used in the stress test.
Another nuance of stress test implementation is that, for purposes
of projecting default and prepayment rates, OFHEO treats all mortgages
with variable payments as if they were standard one-year Treasury ARMs,
with identical payment caps and interest rate margins. In contrast, in
the statistical analysis, specific payment changes for each loan type
were reflected in the creation of explanatory variables.
In the development of explanatory variables for both the
statistical analysis and stress test implementation, a shortcut is used
to amortize ARMs. At each payment adjustment date, the new mortgage
payments are computed using updated interest rates but with the
original UPB and loan term, rather than current UPB and remaining term.
This is seen in the formula used for PMTq, which is the same
for both fixed- and adjustable-rate mortgages. (See section 3.5.2.3,
Procedures of the Appendix.) This approach provides an approximation
for actual payment changes on adjustable rate mortgages. It expedites
calculations by reducing the code necessary to update payments and UPB
in each quarter. The approximation here should have little effect on
default rate results because of the use of categorical, rather than
continuous explanatory variables. Differences in loan amortization
arising from using this payment-calculation approximation only affect
default or prepayment rates when those differences move the probability
of negative equity variable from one (value) category to another. Loan
amortization in the Cash Flow component of the stress test does not use
this shortcut.
In the development of variables for both the statistical analysis
and stress test implementation, the incorrect term is used to amortize
balloon loans. Mortgage origination term (T0), rather than
mortgage amortization term (Ta), is used to amortize these
loans. This is seen in the formula used for PMTq, which does
not distinguish between balloon loans and other loan products. See
section 3.5.2.3, Procedures of the Appendix. Amortization of balloon
loan products in the Cash Flow component of the stress test uses the
mortgage amortization term.
8. Consistency With the Historical Benchmark Experience
Certain adjustments and assumptions to the models were made to
assure consistency of the rates of default projected in the stress test
with the BLE. Loan-level data from the benchmark was aggregated in the
same way current Enterprise loan groups are formed in the stress test,
and the 30-year FRM model was applied to these data to project
conditional and cumulative default and prepayment rates for the ten
years following origination.\229\ A single set of house price
appreciation rates from the OFHEO HPI, the ten-year sequence of
appreciation rates from the West South Central Census division for the
period from 1984 Q1 to 1993 Q4, was applied to every benchmark loan
group.\230\ Actual historical interest rates were used. The projected
average ten-year cumulative default rate was compared to that observed
for the BLE, and adjustments were made to the constant term
D of the default function until the projected and
observed default rates were equal.\231\
---------------------------------------------------------------------------
\229\ Note that all loans of the BLE are newly originated loans.
\230\ The West South Central Census Division does not exactly
match the 4-State benchmark region, but its use here to represent
benchmark economics is consistent with OFHEO's proposal to aggregate
data based on Census divisions, and to apply historical Census
division-level house price growth rates to season loans at the
beginning of the stress test. What is most important is that the
price series used to calibrate the statistical equations is the same
series that will be used in the stress test itself. The actual ten-
year house-price experience of the West South Central Division and
the 4-State benchmark area, 1984-1993, are very similar.
\231\ When computing the cumulative default rate projected by
the model for comparison with that observed for the benchmark
experience, the same calculations were used. The model was used to
project the total defaulting UPB for benchmark loans over the ten-
year period following origination for each monthly origination
cohort. The total defaulting UPB for each Enterprise was obtained by
summing up the total defaulting UPB for each origination cohort,
which was divided by the total original UPB for that Enterprise to
compute the ten-year cumulative default rate. The two Enterprise
cumulative default rates were then averaged. As discussed in NPR1,
because of missing data on defaulting loans, OFHEO used the original
UPBs on default loans in place of UPB at the time of default. This
has little effect on the resulting historical loss rates, because
the same values for defaulting UPBs were used when computing
severity rates. In the calibration of default rates, the UPBs at the
time of default projected from the model (which take into account
normal amortization) were adjusted back to their origination values
for consistency with the benchmark methodology.
---------------------------------------------------------------------------
[[Page 18182]]
The adjusted (calibrated) model is then applied in the stress test,
along with the sequence of house price appreciation rates used in the
calibration procedure.\232\ Therefore, if newly originated loans with
characteristics similar to those comprising the benchmark sample were
subjected to the same economic circumstances as occurred in the
benchmark experience, then the statistical model of mortgage
performance would project ten-year cumulative default rates equal to
those of the benchmark sample. Conversely, to the extent interest
rates, property values, and loan characteristics are different from the
benchmark sample, and to the extent adjustments are necessary to
account for other statutory requirements (e.g., increased general
inflation under large increases in the ten-year CMT), the stress test
rates differ from the benchmark level.
---------------------------------------------------------------------------
\232\ In the calibration, all loans of the BLE are assigned an
HPI volatility parameter estimate based on the West South Central
Census division. In the stress test, loans from each region retain
their respective regional volatility values.
---------------------------------------------------------------------------
The adjustment of the model is appropriate for use in the stress
test because the statistical equations in the model were estimated
using Enterprise data on loans from a broad range of times and places,
in addition to those loans included in the benchmark sample. Because,
by definition, the BLE reflects the highest rates of loss observed from
among these other periods and places, the model would not be likely to
replicate benchmark results on benchmark loans exactly without some
type of adjustment.
The calibration procedure does not add an adjustment factor to
match projected prepayment rates directly to the benchmark prepayment
experience. Nevertheless, the stress test model is fully calibrated to
the credit loss experience of the benchmark loans because the
calibrated default equation, and the uncalibrated prepayment equation
that was used to help calibrate the default equation, are used together
to determine mortgage performance. Because the time paths of Treasury
yields and mortgage rates used in the calibration were those
corresponding to the individual benchmark origination cohorts, the
conditions leading to prepayments in the calibration exercise are
entirely consistent with the benchmark default experience.
[[Page 18183]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.216
[[Page 18184]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.217
[[Page 18185]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.218
[[Page 18186]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.219
[[Page 18187]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.220
9. References
Bartholomew, L., J. Berk, and R. Roll (1988). ``Adjustable Rate
Mortgages: Prepayment Behavior,'' Housing Finance Review, 7:31-46.
Brennan, M. J., and E. S. Schwartz (1985). ``Determinants Of GNMA
Mortgage Prices,'' AREUEA Journal, 13:209-228.
Buser, S. A., and P. H. Hendershott (1984). ``Pricing Default Free
[[Page 18188]]
Mortgages,'' Housing Finance Review, 3:405-429.
Calhoun, C.A. (1996). ``OFHEO House Price Indexes: Technical
Description,'' Washington, D.C., Office of Federal Housing Enterprise
Oversight, April 1996.
Campbell, T.S. and J.K. Dietrich (1983). ``The Determinants of Default
on Insured Conventional Residential Mortgage Loans,'' Journal of
Finance, 38:1569-1581.
Case, K.E. and Shiller, R.J. (1987). ``Prices of Single Family Real
Estate Prices,'' New England Economic Review. 45-56.
Case, K.E. and Shiller, R.J. (1989). ``The Efficiency of the Market for
Single-Family Homes,'' The American Economic Review, 79, 125-137.
Costlett, S. (1981). ``Efficient Estimation of Discrete-Choice
Models,'' pp. 51-111 in C.F. Manski and D. McFadden (eds.), Structural
Analysis of Discrete Data with Econometric Applications, Cambridge,
Massachusetts: MIT Press, 1981.
Cunningham, D., and C. Capone (1990). ``The Relative Termination
Experience Of Adjustable To Fixed-Rate Mortgages,'' The Journal of
Finance, 45(5):1687-1703.
Deng, Y. (1997). ``Mortgage Termination: An Empirical Hazard Model With
Stochastic Term Structure,'' The Journal of Real Estate Finance and
Economics, 14(3), forthcoming.
Deng, Y., J.M. Quigley and R. Van Order (1996). ``Mortgage Default And
Low Downpayment Loans: The Costs Of Public Subsidy,'' Journal of
Regional Science and Urban Economics, 26(3-4):263-285.
Dunn, K.B. And J.J. McConnell (1981). ``Valuation Of Mortgage-Backed
Securities,'' The Journal of Finance, 36:599-617.
Foster, C., and R. Van Order (1984). ``An Option-Based Model of
Mortgage Default,'' Housing Finance Review, 3(4):351-372.
Foster, C. and R. Van Order (1985). ``FHA Terminations: A Prelude to
Rational Mortgage Pricing,'' AREUEA Journal, 13(3):273-291.
Hendershott, P. H. And R. Van Order (1987). ``Pricing Mortgages: An
Interpretation Of The Models And Results,'' Journal of Financial
Services Research, 1:77-111.
Kau, J.B., D.C. Keenan, W. J. Muller, and J. F. Epperson (1985).
``Rational Pricing Of Adjustable Rate Mortgages,'' AREUEA Journal,
13(2):117-128.
Kau, J.B., D.C. Keenan, W. J. Muller, and J. F. Epperson (1990). ``The
Valuation And Analysis Of Adjustable Rate Mortgages,'' Management
Science, 36(12):1417-1431.
Lancaster, T., The Econometric Analysis of Transition Data, New York:
Cambridge University Press, 1990.
Manski, C.F. and D. McFadden (1981), ``Alternative Estimators and
Sample Designs for Discrete Choice Analysis,'' pp. 2-50 in C.F. Manski
and D. McFadden (eds.), Structural Analysis of Discrete Data with
Econometric Applications, MIT Press, 1981.
McFadden, D. (1976), ``Quantal Choice Analysis: A Survey,'' Annals of
Economic and Social Measurement, 5:363-390, 1976.
Price Waterhouse. An Actuarial Review of the Federal Housing
Administration's Mutual Mortgage Insurance Fund, Washington, DC: Price
Waterhouse, 1990.
Richard, S.F. and R. Roll (1989). ``Prepayments on Fixed Rate Mortgage
Backed Securities,'' Journal of Portfolio Management, 15(3):73-82.
Schwartz, E.S. And W.N. Torous (1989). ``Prepayment And The Valuation
Of Mortgage-Backed Securities,'' The Journal of Finance, 44(2):375-392.
Vandell, K.D. ``Handing Over the Keys: A Perspective on Mortgage
Default Research,'' AREUEA Journal, 21(3):211-246.
Zorn, P., and M. Lea (1986). ``Adjustable Rate Mortgage, Fluctuations
In The Economic Environment And Lender Portfolio Change,'' AREUEA
Journal, 14:432-447.
C. Single Family Loss Severity
1. Introduction
This supplementary material provides information on the estimation
and application of statistical models for the single family loss
severity component of the proposed risk-based capital stress test and
regulation. With one exception, all cost and revenue elements of loss
severity are calculated as averages of historical Enterprise experience
with foreclosed mortgages. The one exception is that a statistical
regression model was developed to project the sale proceeds on
foreclosed (real estate owned, or REO) properties. This regression
model uses the same property valuation process that was used to create
a probability of negative equity variable in the default/prepayment
analysis. However, in projecting REO sales proceeds, the process is
used to create a variable that measures the average equity of
performing loans that have the same characteristics (other than equity)
as defaulting loans. The regression then describes the relationship
between average equity of performing loans and average (negative)
equity of defaulting loans. One minus the projected negative equity on
defaulting loans gives the projected REO sale proceeds. This regression
analysis allows stress test loss severity rates to reflect economic
conditions and provides an opportunity to reasonably relate loss
severities on current Enterprise portfolios to the benchmark
experience.
With the exception of government insured loans, OFHEO's loss
severity analysis does not make explicit distinctions by loan product
type. Differences by loan products are captured in the basic loan
terms--coupon rate, LTV, and amortization term-that factor into loss
severity equations.
The Enterprises rely upon various counterparties to provide credit
enhancements that offset gross severity rates. An explanation of how
credit enhancements are modeled in the stress test can be found in the
appendix to the regulation.
The remainder of this supplementary material is organized as
follows: section 2 provides the conceptual framework for single family
loss severity analysis; section 3 describes the data used in the
analysis; section 4 discusses the statistical analysis; section 5
examines adjustments made to the severity equations to reasonably
relate the results to the historical benchmark experience identified in
the first NPR; and section 6 explains how the results of the
statistical analysis are applied in the stress test.
2. Conceptual Framework
In determining the approach to use in modeling loss severity rates,
OFHEO reviewed four research studies. None of these attempted to
analyze the various components of loss severity, but rather used simple
regressions of some measure of a gross severity rate on original loan-
to-value and loan age. These studies provide little guidance, as they
do not provide frequency distributions of observed severity rates, nor
do they provide averages y loan types.\233\
---------------------------------------------------------------------------
\233\ These studies are: Clauretie (1990), Lekkas, Quigley, and
Van Order (1993), Crawford and Rosenblatt (1995), and Berkovec, et
al. (1997). The Berkovec, et al. study is not focused on loss
severities, but rather analyzes them as part of a broader study of
potential lending discrimination. These four studies are reviewed by
Capone and Deng (1998), who themselves are interested in variations
in loss severity rates across defaulted loans that can be explained
by the tenets of option pricing theory. See also Kau and Keenan
(1997) for the one example of severity analysis in a theoretical
mortgage pricing model.
---------------------------------------------------------------------------
OFHEO chose to analyze defaulted loan severity rates in three
parts: loss of loan principal, transaction costs, and
[[Page 18189]]
funding cost. This decomposition was used for three reasons. First, the
loss of unpaid principal loan balance (UPB) is a function of the loss
of property value before and during the default period, which can be
statistically modeled as a function of economic conditions. The second
reason for a decomposition analysis is to accommodate the timing of
various cash flows during the period between initial default (month of
first missed payment) and final property disposition. In the stress
test, all default losses are accounted for in the month of default. The
loss severity rate accounts for the timing of income and expenses after
the default month. The timing of post-default cash flows is captured
using present value discounting techniques. This method also captures
funding costs of the nonearning assets-first the mortgage, and then the
REO. Finally, the stress test calibrates the severity component related
to loss of principal balance to the economic conditions of the BLE, as
will be discussed in section 5. The stress test also uses BLE data for
the elapsed time between default and foreclosure completion, and
between foreclosure completion and property disposition.
Loss severity is most frequently expressed as a rate rather than a
dollar amount. The most accurate representation of the magnitude of
losses is to express loss severity as a percentage of the UPB at the
time of default. Therefore, OFHEO has chosen to calculate all costs and
revenues associated with loss severity as a percentage of the UPB. This
will result in the computation of loss severity rates rather than
dollar amounts, but they become dollar amounts when the stress test
multiplies both default and loss severity rates against loan balances.
3. Data
Loan level data on Enterprise single family REO properties were
used to analyze the components of single family loss severity rates.
The data contain all defaulted mortgages on single family (1-4 unit)
properties that were both originated and had a last-paid-installment
date between January 1980 and December 1995. After removing incomplete
records, over 116,500 valid records remained in the analysis database.
These records consist of loan terms, event dates (default, foreclosure,
disposition), and various expense and revenue fields.
A second analysis database was created consisting of only those
loans in the historical REO analysis database that met benchmark
criteria. Those criteria singled out conventional, 30-year fixed-rate
loans on single family properties (single unit, owner-occupied,
detached properties) that originated in 1983 and 1984 in the States of
Arkansas, Louisiana, Mississippi, and Oklahoma, and defaulted within
ten years of origination. This benchmark database (789 loans) was used
to create an adjustment factor that provides consistency between the
loss severity rates projected in the stress test and the benchmark loss
rates. This process is discussed in section 5, Consistency with the
Benchmark Loss Experience, below.
Other data used in the analysis of loss severity rates includes
historical Census division level HPI indices and their associated
volatility parameters, which come from the OFHEO HPI Report, 1996:3.
4. Statistical Analysis
The primary statistical analysis performed for single family loss
severity rates measured the impact of market conditions on REO sale
proceeds. This is the one dynamic element of loss severity in stress
test application. It relies upon original LTV, loan amortization, and
Census division level house price growth. OFHEO performed a statistical
regression analysis to model negative equity for defaulted loans as a
function of the average equity of similar, but performing, loans. All
other statistical analyses involved calculating average historical
experience by loss severity element. The two elements with values
computed as historical averages are foreclosure expenses and a
combination of REO expenses, revenues (other than disposition
proceeds), and property selling expenses. In addition, average times to
foreclosure and time in REO were computed for use in calculating the
net present value of revenues and expenses in the month of default.
When averages were computed for loss elements, a two-step procedure
was used. First, the average experience of each firm was calculated
using UPB as a weighting factor. This weighted average provides a good
measure of portfolio-wide performance, although the analysis is based
on individual loans. The second step was to give equal weight to the
experience of each firm by taking a simple average of the experience of
the two Enterprises. This procedure is also consistent with the
procedure used to find the benchmark loss severity rate reported in
NPR1.\234\
---------------------------------------------------------------------------
\234\ See 61 FR 29592, 29597, June 11, 1996. Procedures here
differ from those of the first NPR by calculating loss severity as a
percentage of the outstanding loan balance at time of default,
rather than a percentage of the original loan balance.
---------------------------------------------------------------------------
The averages of the foreclosure and the REO expense/revenue
elements are based on the entire national, historical sample of
Enterprise experience. Benchmark experience was not used by itself
because it was evident from an analysis of the data that there were
significant numbers of records with missing expense components. The
magnitudes of these expense items should not vary between the benchmark
region and other areas of the country for two reasons. First, the
benchmark region has a variety of foreclosure laws, by State, so that
the average foreclosure expense rate for the benchmark region is
similar to averages from other regions of the country, and to the
average for the nation as a whole. Second, OFHEO computed these loss
components as percentages of the outstanding loan balance, rather than
as actual dollar amounts. Thus, the fact that the benchmark region may
have had lower property values than the national average, and therefore
lower dollar losses per loan, will not be material. Average loss rate
components from other regions of the country should be comparable to
what would be found in the benchmark loan data, if those records were
complete.
OFHEO does, however, base time frames on benchmark experience.
Because the benchmark region does have a variety of foreclosure laws,
these time frames are actually very close to those of the entire
national experience of the Enterprises.
a. Predicting REO Sale Proceeds
The REO sale proceeds, as a percentage of the defaulting UPB,
measures the impact of erosion of property value over time, both prior
to and after default. To begin the analysis of REO sale proceeds, OFHEO
computed negative property equity, the difference between the
defaulting UPB and the gross property sale proceeds, as a percentage of
the UPB.\235\ This amount was regressed against average equity for
similar, but non-defaulting loans. The resulting regression coefficient
provides the relationship between average equity of performing loans
and average (negative) equity of defaulting loans. The nuance here is
that average equity of performing loans is first transformed into a
standardized normal distance, or what is commonly called a z-score,
before being used in the regression. This is a widely used statistical
technique for
[[Page 18190]]
creating a standard unit of measure for comparisons across many
different variables and/or value levels.
---------------------------------------------------------------------------
\235\ The one expense that OFHEO does net from sale proceeds
here is property repairs undertaken by the Enterprises during the
REO period. Because these expenses reflect part of the loss of
property value that occurred prior to foreclosure completion, it is
appropriate that they be included in the estimation of the loss of
UPB due to property value deterioration.
---------------------------------------------------------------------------
To measure average (performing loan) equity, the property value
underlying each defaulting mortgage was adjusted using the change in
the (Census division) OFHEO HPI from origination to the last-paid-
installment date, and using loan amortization schedules.\236\ This
adjustment provides average expected equity for each loan, if it were
performing. But these loans are not performing, and rather than having
average house price growth, they will generally have lower-than-average
house price growth. In fact, defaulting loans come from the lower tail
of the equity distribution, so the statistical analysis must capture
just how far into the tail defaulting loan properties will be, on
average. OFHEO analyzed several measures of the house price
distribution to find which gave the best prediction of the difference
between average performing loan equity and average non-performing loan
equity. The best predictor was the z-score, identifying the distance
between the expected (performing loan) house price and the (actual
defaulting) loan balance. The z-score transforms the actual difference
between (expected) house price and (actual) loan balance into the
number of standard deviations there are between the two values, where
the standard deviation is of house prices in the Census division. The
z-score tells how far below the average property value growth in the
Census division must the growth of any individual property value be,
before all borrower equity is eliminated. The difference of actual
growth of defaulting loans from average growth for performing loans
will be larger than this, on average, because the z-score distance
gives the minimal difference needed to eliminate borrower equity. The
z-score equation is:
---------------------------------------------------------------------------
\236\ The last-paid-installment (LPI) month is the month
directly prior to the month of default, when the first payment is
missed. Loan amortization ends at LPI, and because the HPI index is
updated quarterly rather than monthly, the choice of LPI month or
default month for loan seasoning is immaterial.
[GRAPHIC] [TIFF OMITTED] TP13AP99.013
[GRAPHIC] [TIFF OMITTED] TP13AP99.227
In their continuous rate forms, the cumulative growth rate factors
are found by taking the logarithm of the HPI, as is done here. The log
of HPI gives average price appreciation, and the difference between
that and the log of the loan balance, B, gives the expected loan equity
due to price appreciation, downpayment, and amortization.\237\
---------------------------------------------------------------------------
\237\ Taking the logarithm of B transforms owner-invested equity
(downpayment plus amortization) into an implied HPI growth rate
factor. It is the cumulative (negative) growth of HPI necessary to
eliminate all positive equity in the property. By transforming B
into its continuous rate counterpart in this fashion, the z-score
variable can measure the amount by which the growth of property
value on loan properties must be less than the average growth rate
of performing loans before default is a real possibility (the point
of zero equity). The regression then measures the relationship
between actual below-normal growth on REO properties and the
minimumly required below-normal house price growth needed to trigger
default.
---------------------------------------------------------------------------
These standardized distances, or z-scores, are the key values used
to compute the expected negative property equity (as a percent of the
outstanding loan balance) when a foreclosed property is sold. Larger z-
scores reflect some combination of large downpayments, loan
amortization, and high levels of (average) house price growth since
loan origination. In these circumstances, loans that do default should
have relatively good rates of property sale proceeds as a percent of
the mortgage UPB (small rates of negative equity). In other
environments, where z-scores are small, there are low rates of
appreciation in the market, and/or low downpayments and a lack of
significant amortization. The small z-score indicates that there is a
wide range of property values in the market area that are below the
loan balance. Therefore, REO sale proceeds will be low and the negative
property equity will be high.
The statistical equation used to predict negative property equity
(L) was estimated using ordinary least squares (OLS) regression of
actual rates of UPB loss on the z-scores computed for each loan. The
regression dataset was limited to historical REO observations where
(-0.50 zt 4.0), because sample sizes
outside this range were very thin.\238\ Log-transformed values of
negative property equity (ln(L) + 1)) were used in the regression to
account for a change in the relationship between negative equity and z-
scores as those values change. The estimated regression equation is:
---------------------------------------------------------------------------
\238\ In stress test application, outliers are given predicted
equity loss values measured at the boundary points of the z-score
range employed in the regression.
[[Page 18191]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.014
[GRAPHIC] [TIFF OMITTED] TP13AP99.228
One-half the regression variance (0.029104) is added to the
regression equation to provide the median-to-mean adjustment factor for
log-normal models.\239\ The result is:
---------------------------------------------------------------------------
\239\ The logarithmic equation used in the regression implies a
lognormal distribution of potential negative equity values around
predicted values. The point estimates from the regression,
therefore, produce median rather than mean value estimates of loss
of principal balance. The adjustment to arrive at the mean is the
additive constant (0.029104), one-half the variance of the
regression residuals.
[GRAPHIC] [TIFF OMITTED] TP13AP99.015
so that:
[GRAPHIC] [TIFF OMITTED] TP13AP99.016
The low R-squared value for the regression indicates a wide
variance of actual loss rates around the average, predicted rates.
OFHEO has analyzed this variance and believes that using the simple
regression equation that captures average loss rates at each z-score
value is more appropriate for the stress test than is a more complex
model that would capture deviations around that average loss rate.
Average rates provide an appropriate simplification because loss
severity rates will be applied to groups of loans.
The boundary values of L are computed at the boundary points of z
used in the regression sample, 4.0 and -0.5. When z = 4.0, L = -0.04.
This suggests that, on average, REO sales prices are 4 percent higher
than the mortgage UPB in areas with significant house price
appreciation and/or for loans that have substantial amortization. That
is, the average default (and there will be relatively few) will
actually have a small amount of positive equity, though generally not
enough to pay the costs of selling the property. At the other extreme,
where z = -0.5, the predicted value of L = 0.36. This is a situation
where average property values on performing loans are 36 percent below
their associated mortgage balances. This extreme was reached in several
areas of the country at various times during the study period. Such a
loss of loan principal can cause the total loss severity to exceed 60
percent of UPB.
b. Foreclosure Expenses
Foreclosure expenses vary principally by property State and by the
rate of bankruptcy filings among defaulted borrowers.\240\ The average
expense rate in the historical observation period is five percent of
UPB. Unlike other loss components, this component is based solely on
Fannie Mae experience because Freddie Mac did not break out foreclosure
expenses from REO expenses in its data systems.
---------------------------------------------------------------------------
\240\ To process foreclosures when defaulting borrowers file for
Bankruptcy Court protection requires further legal expenses to gain
release from the bankruptcy ``stay'' on debt collection actions.
---------------------------------------------------------------------------
c. REO Holding and Disposition Expenses
Property (REO) holding costs include such items as property
maintenance, utilities, property taxes, and hazard insurance. OFHEO
calculated the average total REO holding expenses, plus selling costs
(principally, realtor fees), less miscellaneous revenues to produce a
final REO expense loss severity factor of 13.7 percent.\241\
---------------------------------------------------------------------------
\241\ As noted earlier, the Freddie Mac foreclosure expense rate
is imputed from the Fannie Mae experience (five percent). Therefore,
the REO holding costs used to create the average rate shown here use
total expense for Freddie Mac less imputed foreclosure expense for
Fannie Mae.
---------------------------------------------------------------------------
d. Time Frames
There are two time frames of interest: time from default to
foreclosure completion, and time from foreclosure completion to
property disposition. A mean expected value for each of the time
periods of interest was calculated from BLE data. The mean benchmark
foreclosure time (period from default to foreclosure) was 13 months.
The mean benchmark REO/property sale time was seven months. These time
frames are used in the stress test to discount the various default-
related cash flows to the month of default.
5. Consistency With the Benchmark Loss Experience
The equation for negative equity of defaulted loans (equation 14)
was estimated on all historical REO experience of the Enterprises.
Using this broad range of data assured that the equation would be
appropriate for loans entering the stress test with a wide range of
loan amortization and cumulative HPI experience. The equation used in
the stress test includes an adjustment that calibrates the results to
the BLE.
The procedure for calibrating equation 16 to the benchmark
experience parallels the procedure used by OFHEO to calibrate the
single family default equations to the BLE. A database of defaulted
loans meeting benchmark criteria was input into the negative equity
equation to compute the projected negative equity, by loan. The z-score
variable values were computed by assuming that all loans originated in
the first quarter of 1984, using the West South Central HPI series, for
purposes of assigning house price appreciation rates. These predicted
rates of negative equity were then averaged by Enterprise, using UPB as
a weighting factor. Finally, a simple average of these Enterprise
averages was computed to arrive at a mean expected value for the
benchmark REO database.
[[Page 18192]]
This final mean rate of negative equity on defaulted loans was then
compared with the actual, historical mean rate across the two firms'
benchmark experience. The average projected rate of negative equity
using equation 16 and this averaging method was 21.30 percent. The
actual historical experience average was 31.64 percent. The difference,
10.34 percent, reflects the nature of the benchmark experience: that
defaulting benchmark loans tended to have larger losses, on average,
than did loans from other regions of the country that experienced the
same housing market conditions. The adjusted negative equity equation
is:
[GRAPHIC] [TIFF OMITTED] TP13AP99.017
Proceeds from REO sale are then computed as one minus the projected
negative property equity for the defaulting loans in each loan group.
6. Application to the Stress Test
Stress test application of loss severities begins with the results
of the statistical analysis of severity components discussed here, but
then adds components for loss of loan principal, servicer claim
payments, mortgage insurance, and seller/servicer recourse. OFHEO's
approach is to account for all default related cash flows at one of
three points in time: 120 days delinquency, foreclosure, and property
disposition. The stress test then calculates the effective loss
severity rate as a net present value of all cash flows, in the month of
loan default. The month of default is one month after the last paid
installment (LPI) date, the month of the first missed payment.
There is a difference in the treatment of sold and retained loans
when computing stress test loss severity rates. For retained loans,
defaulting UPB is not a cash outlay and, therefore, is not discounted.
For sold loans, however, the defaulting UPB represents the current
expense of repurchasing a defaulted loan from a security pool. It is,
therefore, a cash-flow element that should be discounted.\242\ This
expense is normally incurred in the fourth month of default. Sold loans
in default also involve four months of interest passthroughs to the
investors while the loans remain in the security pools. The interest
passthroughs are not immediate expenses of the Enterprises because they
are initially matched by passthroughs made by the seller/servicers to
the Enterprises. However, all post-default interest payments received
by the Enterprises are reimbursed to servicers in the post-foreclosure
claim filing. Therefore, all interest passthroughs between seller/
servicers and Enterprises are ignored. Only the passthrough by the
Enterprise to security holders is counted as an expense in the stress
test, and it is included with the seller/servicer claim payment at time
of foreclosure.
---------------------------------------------------------------------------
\242\ Such loans become part of the Enterprise retained
portfolios once they are bought out of the security pools.
---------------------------------------------------------------------------
The stress test provides that, at the time of foreclosure, the
Enterprises make servicers whole for expenses incurred on the loan and
property, including foreclosure costs, and receive proceeds from any
available mortgage insurance. When mortgage insurance is present,
mortgage insurance payments will generally be larger than the servicer
claim payment and provide net inflows of funds to the Enterprises at
foreclosure.
Also, any available seller/servicer recourse is applied to reduce
the final loss severity rate. There are some smaller sources of credit
enhancements that further reduce Enterprise losses, and these are added
once dollar losses are computed in the cash flow component of the
stress test.\243\
---------------------------------------------------------------------------
\243\ These lesser sources of credit enhancements are items
where the amount of recourse available to the Enterprises is not a
function of per loan losses, but rather it is available in total
dollar amounts for pools of loans.
---------------------------------------------------------------------------
7. References
Berkovec, James A., Glenn B. Canner, Stuart A. Gabriel, and Timothy
Hannan. 1998. Discrimination, Competition, and Loan Performance in FHA
Mortgage Lending, Review of Economics and Statistics, forthcoming.
Capone, Charles A. and Yongheng Deng. 1998. ``Loss Severities and
Optimal Put Exercise: An Examination of Negative Equity in Mortgage
Foreclosure,'' unpublished manuscript. OFHEO: Washington, DC, January
1988.
Clauretie, Terrence. 1990. ``A Note on Mortgage Risk: Default vs. Loss
Rates,'' AREUEA Journal 18 (2), 202-206.
Crawford, Gordon and Rosenblatt, Eric. 1995. ``Efficient Mortgage
Default Option Exercise: Evidence from Loss Severity,'' Journal of Real
Estate Research 19 (5), 543-555.
Kau, James B. and Donald C. Keenan. 1993. ``Transaction Costs,
Suboptimal Termination, and Default Probabilities for Mortgages,''
AREUEA Journal 21(3), 247-63.
Kau, James B. and Donald C. Keenan. 1997. Patterns of Rational Default,
unpublished working paper, University of Georgia.
Lekkas, Vassilis, John M. Quigley and Robert Van Order. 1993. ``Loan
Loss Severity and Optimal Mortgage Default,'' AREUEA Journal 21 (4,
Winter), 353-372.
D. Multifamily Default/Prepayment
1. Introduction and Conceptual Framework
This section describes how OFHEO developed its model of multifamily
default and prepayment rates for use in the risk-based capital stress
test. The same theory that underlies the single family default/
prepayment models, financial options theory, also underlies OFHEO's
modeling of mortgage performance for multifamily loans. However, the
single family approach is modified to account for the importance of
property cash flows in the default decisions of investors. This
theoretical framework treats mortgage terminations as a function of
their financial value to the borrower. Both the single family and
multifamily default/prepayment models also use a multinomial logistic
specification to estimate the impact of explanatory variables on
default and prepayment rates. Beyond these similarities in general
approach, however, there are significant differences in the specifics
of model construction and estimation.
Many of these differences reflect special features of multifamily
mortgages. For these loans, the borrowers are all investors, and that
affects the determinants of credit risk. Two key financial ratios are
used in commercial mortgage underwriting: the DCR and the LTV. DCR is a
property's net operating income (NOI) divided by the mortgage
payment.\244\ DCR indicates how much cash there is available for loan
repayment after operating expenses are paid. LTV is the ratio of the
UPB to the value of the property; it measures
[[Page 18193]]
borrower equity.\245\ Lenders concentrate on these two ratios at loan
underwriting, and all major credit rating agencies start their analysis
of the credit support levels needed to receive various rating grades
with the DCR and LTV values of the loan collateral.
---------------------------------------------------------------------------
\244\ NOI is a measure of the differernce between full potential
rent at market prices and operating expenses (including vacancy
losses).
\245\ Commercial loan underwriting also includes examinations of
borrower credit, servicing capability, site and engineering reviews,
and cost certifications for new construction. Market condition
reports are part of the appraisal process used to estimate LTV at
loan origination.
---------------------------------------------------------------------------
Multifamily mortgage modeling should also recognize the special
features that differentiate commercial loans from single family
residential loans. Commercial loans have prepayment restrictions,
usually in the form of yield maintenance clauses, that severely reduce
the value of refinancing during the early years of a mortgage.
Commercial loans are also dominated not by fully amortizing 30-year
loans, but by balloon mortgages with maturities of up to 15 years.
These two product distinctions--yield maintenance and balloon terms--
create different borrower incentives and different mortgage performance
patterns for multifamily mortgages.
Previous research on multifamily mortgage performance has generally
made simplifying assumptions to avoid having to deal with all of these
issues in one model. First, research has tended to ignore DCR and only
concentrate on LTV. Even then, without readily available property value
indexes, researchers have not updated LTV over time to capture local
market conditions.\246\ Some studies have captured property cash flows,
but they omitted LTV and had no mechanism for updating property cash
flows for projection purposes.\247\ One study that recognized the need
for both DCR and LTV for predicting default rates, defined them to be
perfectly correlated so that only one financial variable needed to be
included in the model.\248\ Another shortcoming of past research has
been that default and prepayment have not been analyzed together.\249\
Either defaults are assumed not to matter because of agency guarantees,
or else prepayments are ignored because of yield maintenance terms.
Most studies model defaults without prepayments, but prepayment studies
are starting to appear, with three in 1997 and one in 1998.\250\ In
both default and prepayment studies, little work has been done to
understand the dynamics of yield maintenance and balloon terms.\251\
But even with all of these limitations in current research, the
greatest concern is that researchers most often resort to pooling
multifamily mortgages with loans on other commercial property types in
order to have sufficient sample sizes.\252\
---------------------------------------------------------------------------
\246\ Vandell (1992) and Vandell, et al. (1993) develop models
of commercial mortgage default that update LTV over time using a
national property-value index, along with the property-value
diffusion process introduced by Foster and Van Order (1984) for
single family mortgages.
\247\ See ICF (1991) and Pedone (1991). These studies adapt the
work of Edward Altman (1981, 1983) to predict corporate bankruptcy
to model multifamily defaults. Capone (1991) discusses the
application of bankruptcy models to multifamily mortgages, and
provides a review of this literature. A related line of literature
discusses the relationship between lender and borrower in the
default/bankruptcy process. Kahn (1991) and Mahue (1991) study the
impact of foreclosure laws on the balance of borrower and lender
bargaining strength at these crucial junctures. Riddiough and Wyatt
(1994a, 1994b) explore the power of lender signals of intent to
pursue debt collections on distressed-loan foreclosure.
\248\ Abraham (1993b).
\249\ The first known attempt outside of OFHEO to model default
and prepayment rates simultaneously was by Boyer, Follain, Ondrich,
and Piccirillo (1997), who studied FHA insured mortgages.
\250\ Abraham and Theobald (1997), Elmer and Haidorfer (1997),
Follain, et al. (1997), and Capone and Goldberg (1998).
\251\ In a theoretical pricing model, Kau, et al. (1990) do
attempt to show how prepayment restrictions impact both default and
prepayment options with balloon mortgages.
\252\ The lack of historical data has often been cited as a
major obstacle to research on multifamily and commercial loan credit
risk (DiPasquale & Cummings, 1992; Standard & Poors, 1993; and
Vandell, et al., 1993). Studies that combine multifamily with other
commercial mortgage types include Vandell (1992), Vandell, et al.
(1993), Barnes and Gilberto (1994). Studies that use only
multifamily data tend to model FHA-insured loans (Goldberg, 1994;
ICF, 1991; Follain, et al., 1997). Exceptions to this include
Abraham (1993a, 1993b), who used multifamily loan data from Freddie
Mac to study defaults, and Abraham and Theobald (1997), who use
Freddie Mac data to model multifamily prepayment rates. Elmer and
Haidorfer (1997) use Resolution Trust Corporation data to study
multifamily prepayment rates. Researchers at OFHEO have published a
default study based on Enterprise data (Goldberg and Capone, 1998).
---------------------------------------------------------------------------
The broad conceptual framework chosen by OFHEO corresponds to the
dominant paradigm in mortgage research, financial options theory.
Studies that apply financial options theory to commercial mortgage
performance have generally emphasized the role of borrower equity (LTV)
in default rate estimation, but have not seriously modeled the role of
cash flows (DCR).\253\ However, because both DCR and LTV are critical
credit risk dimensions, an appropriate multifamily mortgage performance
model should also treat cash flows and equity as essential
elements.\254\
---------------------------------------------------------------------------
\253\ Even theoretical ``pricing'' models that simulate default
rates on a pool of newly originated mortgages make simple
assumptions that cash flow to the property owner is a fixed
percentage of property value (Titman and Torous, 1989; Kau, Keenan,
Epperson, and Muller, 1987 and 1990). They also treat cash flow as
something negative (detracts from potential future property value)
rather than something positive to the investor/owner/borrower.
\254\ Abraham (1993b), Goldberg (1994), and Quercia (1995) have
all questioned the sufficiency of net equity as a default trigger.
---------------------------------------------------------------------------
For the default option to be in the money, the property must have
both negative equity (LTV>1) and negative cash flow (DCR<1). The two
sources of income for an investment property owner are rental (current)
income and capital gains. Rental income can be thought of as dividend
payouts from the property. Capital gains result when the property is
sold. The owner holds the property until the expected annual rate of
return from both dividends and capital gains becomes less than the
return that could be earned by selling the property and investing the
proceeds into another investment. However, if the rental market
declines, and property equity becomes negative, then default becomes a
viable option. This option will not be exercised as long as the
dividend payout is positive. If property owners/borrowers were to
default in the presence of positive cash flows, they would give up
valuable cash flow streams. Therefore, default is only optimal if both
equity and cash flow are negative. This implies that the dual
condition, LTV> 1 and DCR<1, is required for default to occur.\255\
---------------------------------------------------------------------------
\255\ The wealth-maximizing borrower should default if the
property expects to have negative equity and negative cash flow from
this point on. If there are negative cash flows, delaying default
would lower wealth. If negative equity and negative cash flow were
expected to be only temporary conditions, default would not be
optimal. In principle one should incorporate expectations regarding
rental markets and interest rates, simulate wealth over time, and
have the borrower default only if it maximizes wealth over some
long-run investment horizon. This was viewed as an overly complex,
expensive, and therefore unfeasible approach. Theory
notwithstanding, researchers typically construct the default option
value variable using just current year information. This is also the
approach taken by OFHEO. For relevant theoretical studies, See Kau
et al. (1987, 1990), Brennan and Schwartz (1985), Dyl and Long
(1969), Joy (1976), and Robichek and VanHorne (1967).
---------------------------------------------------------------------------
Prepayment options are in some ways simpler and in others more
complex than default options. The simplicity arises because the
financial value of prepaying a mortgage is directly measured by the
mortgage premium value, the difference between the present value of
future mortgage payments discounted at the current note rate, and
present value of those same payments discounted at the current market
rate. When interest rates fall, there is negative value to holding onto
the existing mortgage, measured by a negative mortgage premium value.
However, measuring the premium value itself is complex because of yield
maintenance and balloon terms. When a
[[Page 18194]]
fixed-rate loan is under yield maintenance, it may refinance, but it
will not accrue any value from the transaction until the yield
maintenance period expires.\256\ With balloon loans, there is the added
uncertainty surrounding the contractual requirement to find new funding
at loan maturity. Risk averse borrowers, therefore, may desire to
refinance in the pre-balloon period even if the call option is not in
the money.
---------------------------------------------------------------------------
\256\ ARM loans have minimal penalties, and they have prepaid
much more often in the early years after loan origination.
---------------------------------------------------------------------------
An additional consideration for modeling prepayment speeds is that
investors desire to leverage their investments to maximize return on
equity. Interest rate spreads do not, therefore, provide the only
incentive for refinancing a mortgage. To maximize leverage requires
maximizing LTV ratios, within bounds set by lenders. Over time,
investors will engage in cash-out refinancings in order to rebalance
the ratio of debt to equity in the property. This second prepayment
incentive can be captured by the LTV of the mortgage.
In modeling multifamily mortgage default rates, OFHEO distinguishes
among the various programs of the Enterprises. Conventional multifamily
loan purchases by the Enterprises began in 1983, and include ``cash''
and ``negotiated'' programs. Under the cash programs, the Enterprises
purchased newly originated individual loans underwritten according to
their own guidelines. Historically, most of these loans were retained
in the portfolios of the Enterprises. Some ``cash'' loans were swapped
for MBS, and this type of transaction is becoming more common. In a
negotiated transaction, an Enterprise swaps pools of seasoned (i.e.,
aged and performing) loans for securities. These loans need not meet
the underwriting guidelines of cash programs, and they are priced
according to the risk of the loans in the pool. In negotiated
transactions, unlike cash purchases, an Enterprise often requires
credit enhancement from the seller/servicer to cover expected credit
losses.
The initial cash programs exposed the Enterprises to significant
credit risk in the late 1980s and into the 1990s. This exposure was due
to generous appraisal practices used in the 1980s and to other
significant weaknesses in those programs that do not exist today.
Fannie Mae changed its cash program in 1988. Freddie Mac continued to
build a portfolio of less-than-investment-grade mortgages through 1990.
The poor performance of this portfolio led to a three-year moratorium
on Freddie Mac's new purchases of multifamily loans, and a complete
overhaul of the multifamily operations of the Enterprise.
Prepayment rates were modeled by loan characteristics product type
rather than program type. This breakdown captures the differences in
financial incentives to prepay that exist when yield maintenance
penalties are or are not in effect, and the impact on defaults of
balloon mortgage maturity. Balloon maturity is a significant
multifamily modeling issue for the stress test because, in an up-rate
interest rate environment, balloon loan borrowers are often required to
pay off the existing mortgage and refinance, at much higher interest
rates than property financials are currently supporting. In order to
refinance at the balloon point in the up-rate scenario, property income
must be higher than the minimum necessary to qualify for a new loan
under the original interest rates. Therefore, it is important to model
both the expected default and payoff rates of loans at balloon maturity
for the stress test.
Section 2 of this supplementary material on multifamily default/
prepayment provides a review of the historical data used to estimate
the statistical models, and section 3 reviews the statistical
procedures employed. Section 4 completes the description of the
statistical model with explanations of the development of the
explanatory variables. Section 5 presents and reviews the results of
statistical estimations, and section 6 concludes with a discussion of
how the estimated statistical equations are applied in the stress test.
2. Historical Data
a. Enterprise Loan Records
OFHEO used the combined historical experience of the Enterprises,
1983-1995, to estimate the statistical model of default and prepayment
rates. This experience provided a large and rich data base that
encompasses three different programs: the initial cash purchase
programs that had high default rates; negotiated purchase (or
transactions) programs where securities were swapped for pools of
seasoned and performing mortgages; and new cash purchase programs that
corrected flaws in the original programs and have experienced low
default rates.
The historical data includes 35,759 conventional multifamily
loans.\257\ After eliminating missing or erroneous records, the sample
includes observations on 21,994 loans: 12,845 from Freddie Mac and
9,149 from Fannie Mae. Of these, 61 percent are cash purchases and 39
percent are negotiated purchases. The final cash purchase sample is
more complete than the negotiated purchase sample because, in
negotiated programs, the Enterprises have relied more on buying
seasoned portfolios with (limited) credit risk recourse to the seller/
servicer, rather than on gathering enough property financial
characteristics to re-underwrite the loans.\258\
---------------------------------------------------------------------------
\257\ Fannie Mae has maintained a portfolio of FHA-insured
multifamily mortgages over time. OFHEO chose not to model
performance of these loans, but rather to assign default and
prepayment rates according to conventional loans with similar
features. Because FHA pays for nearly 100 percent of default losses,
the stress test imposes no credit losses on FHA-insured mortgages on
the stress test.
\258\ Ninety percent of cash purchases are retained in the final
sample, while only 41 percent of negotiated purchases had enough
loan characteristics data to be kept in the sample. For the 41
percent of negotiated purchase loans in the sample, DCR values at
time of acquisition were estimated by OFHEO by first estimating net
operating income (NOI) as NOI=value at origination divided by an
estimate of the average CAP rate multiplier for the year, divided by
the mortgage payment amount.
---------------------------------------------------------------------------
The database was expanded by creating annual observations from loan
acquisition to the termination year, or to 1995 if no termination
occurred. The loan-year file includes 89,577 loan-year observations for
cash purchases, and 59,415 observations for negotiated purchases. Cash
purchases appear in the database with origination years from 1983 to
1995. The negotiated loans, however, have origination years as early as
1970 because they were often highly seasoned at time of acquisition.
Annual observations are used, rather than monthly or quarterly
observations, because of the relatively small number of multifamily
termination events. If quarterly or monthly event histories were used,
there would be significant numbers of time periods in which there were
no terminations.
To avoid any possible statistical bias resulting from not having
records of loan terminations prior to 1983, negotiated purchase loans
enter the database starting in the acquisition year, rather than the
origination year. But they enter at their proper age and are not
treated as new originations at the time of acquisition. The same issue
of potential ``left censoring'' bias also appears for certain cash
purchase programs, where the Enterprises did not begin to maintain
systematic records of loan terminations until 1991. For such programs,
the loans do not enter the statistical estimation sample until
1991.\259\
---------------------------------------------------------------------------
\259\ The left-censoring bias would result if the statistical
model used complete loan-history records for all loans, when some
groups of loans only enter the sample if they survive to a certain
point (e.g., time of acquisition by the Enterprise). If the sample
were not censored at the acquisition point, the model could severely
underestimate the rates of loan termination in the early years of a
mortgage.
---------------------------------------------------------------------------
[[Page 18195]]
For cash loans, the default outcome of record is a foreclosure or
foreclosure alternative that still provides for the property to be
liquidated.\260\ For most Fannie Mae negotiated purchase loans,
however, the default event of record is a 90-day delinquency. This is
because, for Fannie Mae negotiated transactions, the loan is
repurchased by the seller/servicer if it becomes 90-days delinquent.
The seller/servicer then bills Fannie Mae for resolution costs, and
these are deducted from a limited recourse pool originally established
with funds from the seller/servicer at time of acquisition. OFHEO
recognizes that 90-day delinquencies cannot be treated as full default
events, and makes adjustments in the statistical model.
---------------------------------------------------------------------------
\260\ Foreclosure alternatives include third party sales where a
``third party'' purchases the property at the foreclosure auction;
short sales, where the Enterprise finds a buyer for the property
prior to completion of foreclosure; and note sales, where the
mortgage itself is sold to another investor.
---------------------------------------------------------------------------
b. Rents and Vacancies
OFHEO uses a unique approach to property valuation that uses local
market indexes of rent growth rates and vacancy rates to update net
operating income, and through that, update DCR and LTV over time. Rent
growth rates came from the residential rent component of the CPI for
each of the four Census regions, and for the 29 MSAs covered by Bureau
of Labor Statistics (BLS) surveys. Most MSA level CPI series produced
by BLS start in 1970, but some do not begin until the 1980s. The
regional CPI series are available beginning in 1978, so percent changes
for these can only be computed starting in 1979. To capture rent growth
rates for each year, partial MSA series were completed with regional
series starting in 1979 and national series before that. The regional
series themselves were also filled in for the pre-1979 period with
percent changes in the national CPI residential rent series.
Vacancy rates were obtained from the Bureau of the Census H-111
series. These are available for the same MSAs as is the CPI residential
rent series (back to 1970), and for Census regions, and, beginning in
1986, for the 50 States plus the District of Columbia.\261\ As with
rent growth rates, the most disaggregated index available was used for
each loan, in each calendar year.
---------------------------------------------------------------------------
\261\ Census also added more MSAs starting in 1986. These were
not used in OFHEO's statistical analysis.
---------------------------------------------------------------------------
c. Tax RatesOFHEO required tax rate data for calculating the
present value of depreciation writeoffs (see discussion of the
explanatory variable, DW, below). In order to compute weighted
average tax rates, OFHEO used Internal Revenue Service (IRS) data
on the income distribution of taxpayers with net capital gains. For
1983-90, data on adjusted gross income for taxpayers with net
capital gains were obtained from the IRS publication, Individual
Income Tax Returns (annuals). For 1991-95, data were obtained from
IRS, Statistics of Income Bulletin (quarterly). These income-class
weights were used to compute weighted average tax rates for both
capital gains and ordinary income.
The marginal tax rate on ordinary income used here is for Married
Filing Jointly taxpayers (Schedule Y-1). Five percent was added to the
Federal tax rate for State income taxes. Schedule Y-1's for 1983-95
were obtained from Internal Revenue Service, Package X (annual
publications 1983-95). Data on capital gains tax rates were obtained
from IRS's Package X, for 1983-95. No adjustment was made for State
taxes on capital gains.
Data on depreciation schedules is for newly constructed residential
rental property, from the IRS publication, Depreciation 1992,
Publication 534. This publication includes accelerated schedules for
years 1983-92. Accelerated depreciation was assumed in years in which
it was an option. Because there were no changes in the tax code
affecting depreciation after 1992, the schedule for 1992 was used for
1993-95.
3. Statistical Estimation
The statistical estimation involves binomial logistic regressions
of subsets of the data. There are two separate regressions for default
rates and five separate regressions for prepayment rates. This
breakdown accommodates programmatic differences between cash and
negotiated purchases in the default equations, and the changing nature
of prepayment incentives across various products and loan terms. The
results are matched together so that the end result is trinomial
logistic probability equations that provide the same result as if
defaults and prepayments were estimated simultaneously for each loan
program and product.\262\
---------------------------------------------------------------------------
\262\ This is the three-choice logit model, though the more
generic model is known as the multinomial logit, or MNL.
---------------------------------------------------------------------------
The logistic model is founded on assumptions that the utility of
each borrower payment choice--make payment, prepay, or default--is a
function of its contribution to wealth and that, each observation
period, borrowers make the choice that maximizes wealth. The
regressions compute weights (coefficients) that estimate the influence
of each explanatory variable on the net wealth effect of one choice
over another. These models estimate the log-odds of choosing a mortgage
termination over continuing to make loan payments as a function of the
explanatory variables. In particular,
[GRAPHIC] [TIFF OMITTED] TP13AP99.018
and
[GRAPHIC] [TIFF OMITTED] TP13AP99.019
[[Page 18196]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.229
And the resulting equations for calculating probabilities are
transformations of these equations:
[GRAPHIC] [TIFF OMITTED] TP13AP99.020
and
[GRAPHIC] [TIFF OMITTED] TP13AP99.021
If X and Y are matrices of all event-history records, then the
resulting probabilities will be (column) vectors of estimated
probabilities for each of these records, for each observed time period.
Because of the relatively small number of loan defaults in the data,
OFHEO used annual observations to estimate the equations. Economic
variables are averages for each calendar year, and the logistic
equations estimate probabilities of default and prepayment for all
loans surviving to the beginning of the next year.
The probabilities of default and prepayment are interdependent, and
normally the equations would be estimated using simultaneous equations
methods. However, because there are two default equations and five
prepayment equations, doing so would be quite complex. Following Begg
and Gray, OFHEO estimated the system using single equation methods in
which separate binomial log-odds equations are estimates for default
and prepayment.\263\
---------------------------------------------------------------------------
\263\ See Begg and Gray (1984). To do this, one must be sure to
censor competing termination events from the regression samples.
That is, for default rate log-odds estimation, all prepayment
observations must be censored in the period of the prepayment (and
vice versa). This censoring assures that the estimation is of the
log-odds of defaulting (or prepaying) versus remaining current on
the mortgage. The underlying principle of logistic regression
analysis that allows for this approach to modeling the competing
risks of default and prepayment is called the independence of
irrelevant alternatives. This principle means that logistic analysis
assumes that the log-odds of default versus remaining current are
not influenced by the log-odds of prepaying versus remaining
current.
---------------------------------------------------------------------------
4. Explanatory Variables
The multifamily mortgage performance model has separate sets of
explanatory variables for default and prepayment analysis. They are
described separately here.
a. Default Equations
OFHEO estimated two separate logit default equations, one for cash
purchases and one for negotiated purchases. This decomposition serves
three purposes. First, significant numbers of negotiated purchase loans
did not enter the Enterprise portfolios until after the Tax Reform Act
of 1986. That statute greatly changed the value of depreciation
allowances to new purchasers of investment real estate. OFHEO desired
to model the effects of tax law changes on default rates, but could
only do this with the cash purchase loans, where there are significant
numbers of observations both before and after tax reform. The second
reason for separating cash from negotiated purchase loans is that
negotiated loans did not undergo the same change of quality as did cash
purchases. It is easier to separate the effects of movements by the
Enterprises from original to new cash-purchase programs if these are
isolated from the negotiated purchases for default analysis. A third
reason for separating the two programs into two separate default
equations is that the majority of negotiated purchase loans have
seller/servicer repurchase provisions, which required use of 90-day
delinquency as the default event of record. OFHEO decided that
capturing the difference between 90-day delinquencies and full defaults
was best achieved through an estimation that involved only negotiated
purchases.
Table 33 provides a list of the explanatory variables used in each
default equation. Each variable listed in the Table will be described
and developed more fully below.
[[Page 18197]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.221
(i) Joint Probability of Negative Equity and Negative Cash Flow
The key explanatory variable in the default equations is the joint
probability of negative equity and negative cash flow, which is defined
as:
[GRAPHIC] [TIFF OMITTED] TP13AP99.022
A probabilistic measure is used because the exact financial
condition of each mortgaged property, over time, is unknown. However,
the equity and cash flow positions of the property at time of loan
acquisition, and how local rents and vacancy rates changed over time
are known. With this information, and reasonable assumptions regarding
the dispersion of rent growth rates and vacancy rates across
properties, the joint probability, JP, can be constructed. This
variable is similar to the probability of negative equity variable used
in the single family mortgage performance model, only here the variable
begins with an index of growth rates of property net operating income
(NOI), rather than an index of the growth rates of property value
directly. OFHEO developed this approach for multifamily modeling
because there are no property value indexes available, and it was not
feasible to develop one with Enterprise data.
Ideally, JP would capture all of the numerous factors affecting LTV
and DCR, including rents, expenses, vacancies, special underwriting
provisions (e.g., maintenance reserves), interest rates, and tax laws.
OFHEO incorporated three important factors into the JP variable: rents,
vacancies and interest rates. Because the actual property purchase year
for current investors is unknown, the actual tax code affecting
depreciation writeoffs is also unknown for each property. Therefore,
OFHEO constructed a separate variable that captures changes in the
value of tax benefits from property ownership to a new purchaser.
Changes in property expenses are incorporated into JP by specifying
that expenses are a constant ratio of rents.
(a) Creating Time Series for DCR and LTV
The construction of JP first involves creating time series
variables for DCR and LTV. Each of these can be shown to be a function
of property NOI in each time period, t:
[GRAPHIC] [TIFF OMITTED] TP13AP99.023
[[Page 18198]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.230
and
[GRAPHIC] [TIFF OMITTED] TP13AP99.024
[GRAPHIC] [TIFF OMITTED] TP13AP99.231
For commercial properties, appraisers use capitalization (``cap'')
rate factors for estimating the present value of a future stream of
property NOI.\264\ The cap rate multiplier for each loan at
origination, M0, can be derived given three other variables:
LTV0, UPB0, and NOI0. Because the cap
rate multiplier is a function of interest rates, changes in interest
rates over time will affect Mt and, through that, affect
Vt and LTVt also. OFHEO collected data on cap
rate multipliers at origination on Enterprise loans and the mortgage
coupon rates on those loans.\265\ These data were used to estimate the
elasticity of the cap rate multiplier with respect to interest rates,
so that property values can be updated in response to interest rate
changes. The estimated regression equation is:
---------------------------------------------------------------------------
\264\ While the cap rate multiplier is used here to project
property value from NOI, the cap rate itself is the reciprocal of
the multiplier. So if, for example, a cap rate multiplier of 10 is
implied from the property value (and the underlying NOI), the actual
cap rate is 0.10. The cap rate on each individual property begins,
like other appraisal techniques, with cap rates found on recent
sales of comparable properties. Appraisers then incorporate an
assessment of the duration and risk of the earnings on the
particular property into the final cap rate used to project property
value; a risky earnings stream will be penalized with a higher cap
rate (lower multiplier).
\265\ The choice of an interest rate series to use here was one
of convenience, and does not materially affect the results.
[GRAPHIC] [TIFF OMITTED] TP13AP99.025
[GRAPHIC] [TIFF OMITTED] TP13AP99.232
By estimating a double-log equation, the coefficient on the
interest rate variable, rc,0, is the elasticity of the cap
rate multiplier with respect to interest rate changes. This elasticity
is used to project changes in Mt over time (since loan
origination) as follows:
[GRAPHIC] [TIFF OMITTED] TP13AP99.026
where:
[GRAPHIC] [TIFF OMITTED] TP13AP99.233
and the factor used to update LTVt over time is then,
[[Page 18199]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.027
Updating the DCRt and LTVt series requires a
method for updating NOIt. NOIt can be expressed
as a function of rents, operating expenses, and vacancy rates:
[GRAPHIC] [TIFF OMITTED] TP13AP99.028
[GRAPHIC] [TIFF OMITTED] TP13AP99.234
For national data from annual surveys of apartments by the
Institute for Real Estate Management (IREM), from 1970 through 1992,
the average ratio of operating expenses to gross rents was 47.2
percent. In computing values for NOIt, kt is held constant
at 0.472. All properties must meet minimum occupancy (maximum vacancy)
requirements before permanent funding is secured and the loans are
purchased by the Enterprises. To estimate the models, the vacancy rate
at origination is also held constant, in this case at the long-term
average observed in Census vacancy surveys, 1970-1995: 0.0623. Thus,
current values of NOI, relative to the value at loan origination, are
calculated as:
[GRAPHIC] [TIFF OMITTED] TP13AP99.029
Which reduces to:
[GRAPHIC] [TIFF OMITTED] TP13AP99.030
The ratio of the dollar rents (RENTt/RENT0)
is a rent index from the time of loan origination to time period t.
Denoting this ratio as RPIt,:
[GRAPHIC] [TIFF OMITTED] TP13AP99.031
If there has been no change in the vacancy rate since loan
origination, so that vt = 0.0623, then the growth of net
operating income equals the growth of rents since loan origination.
With these formulas for updating the cap rate multiplier,
Mt, and net operating income, NOIt, time series
can be created for DCRt and LTVt.
[GRAPHIC] [TIFF OMITTED] TP13AP99.032
and
[GRAPHIC] [TIFF OMITTED] TP13AP99.033
[[Page 18200]]
For DCRt, the denominator of the update formula
(PMTt/PMT0) (equation 32) will always equal 1.0
for fixed-rate mortgages.
(ii) Construction of the JPt Variable
Both RPIt and t are market indexes.
The values for individual properties--RPIj,t and
j,t--are not known, but can be assumed
to be random variables that follow standard distributional forms, with
mean values RPIt and t. To look at how
the distribution of rent growth rates and vacancy rates affects the
distribution of property level DCR and LTV values, it is convenient to
use a logarithmic transformation of equation (31):
[GRAPHIC] [TIFF OMITTED] TP13AP99.034
where Zt = [1-2.15 (t-0.0623)] and
RPIt is a rent index that equals one plus the growth of
rents since loan origination. Zt can be interpreted as the
percentage change in NOIt due to changes in the vacancy rate
since loan origination, and RPIt is the percentage change in
NOIt due to rent growth. If ln(Z) and ln(RPI) are normally
distributed across properties, at any given point in time, then their
sum has a bivariate normal distribution. This implies a bivariate
normal distribution for ln(DCR) and ln(LTV), which provides the
distributional form used to estimate the joint probability that DCR < 1
and LTVt > 1 for any given property, JPt.
Normality for ln(RPI) follows from the standard assumption that
growth rates follow a lognormal diffusion process over time. Such a
process is also foundational to the OFHEO HPI, which is used for single
family mortgage performance analysis. With lognormal diffusion, the
distribution of ln(RPIj,t), where j is a property index, is:
[GRAPHIC] [TIFF OMITTED] TP13AP99.035
[GRAPHIC] [TIFF OMITTED] TP13AP99.235
If all apartment units can be assumed to have the same probability
of being vacant, the distribution of vacancy rates across properties,
within a geographic area, can be assumed to be binomial, with mean and
variance parameters:
[GRAPHIC] [TIFF OMITTED] TP13AP99.036
[GRAPHIC] [TIFF OMITTED] TP13AP99.236
The binomial distribution for apartment vacancies at the project
level is bounded below by zero and skewed to the right, and because it
can be approximated by a lognormal distribution with the same
parameters. Thus, Zj,t, which is a linear transformation of
vj,t, can be modeled with a lognormal distribution:
[GRAPHIC] [TIFF OMITTED] TP13AP99.037
This allows ln(Zj,t) to be modeled with a normal
distribution. Rewriting the parameters of Zj,t as:
[GRAPHIC] [TIFF OMITTED] TP13AP99.038
we can write the parameters of the (normal) distribution of
ln(Zj,t) as:
[GRAPHIC] [TIFF OMITTED] TP13AP99.039
[[Page 18201]]
where the t subscripts for these parameters are dropped here and
subsequently for clarity. Because both ln(DCRj,t) and
ln(LTVj,t) are linear functions of the normally distributed
random variables, ln(Zj,t) and ln(RPIj,t),
ln(DCRj,t) and ln(LTVj,t) have a bivariate normal
distribution,
BV(1,2,1,
2,), where,
[GRAPHIC] [TIFF OMITTED] TP13AP99.040
The correlation between ln(LTVj,t) and
ln(DCRj,t) in the historical Enterprise data is used as an
estimate of (-0.5975). Unpublished data from the Bureau of
Labor Statistics (BLS) suggests a value for
2t of 7.5 percent. Alternative values
between 5 and 15 percent were also considered, but the statistical
model results (default rate equations) were insensitive to the value
used for this variance.\266\
---------------------------------------------------------------------------
\266\ This is because the variance of lnDCR and lnLTV is much
more heavily influenced by the variance of the vacancy rate than the
variance of the growth rate of RPI.
---------------------------------------------------------------------------
The bivariate normal distribution defined by the parameters in
equation 40 can be used to calculate the joint probability of negative
equity and negative cash flow, JP. The joint probability is the
bivariate (standard) normal distribution evaluated at particular
boundary (cutoff) values for ln(DCR) and ln(LTV). The definition of
JPj,t can be restated as:
[GRAPHIC] [TIFF OMITTED] TP13AP99.041
which can be calculated using the bivarate normal distribution:
[GRAPHIC] [TIFF OMITTED] TP13AP99.042
where x and y are two standard normal random variates, each
representing the possible values of the logs of DCR and LTV values on
all apartment properties in a given geographic area, at a given point
in time. The x and y values are standardized, which for DCR and LTV is
accomplished by subtracting from them the log of the expected values
for each property, 1,t and 2,t,
and then dividing by the respective standard deviations,
1,t and 2,t. The two limits of
integration, a and b, are the standardized differences between the
expected values for each property and the boundary conditions, which
are the log of 1.00 for each. So, from equation 40 they are just:
[GRAPHIC] [TIFF OMITTED] TP13AP99.043
and
[GRAPHIC] [TIFF OMITTED] TP13AP99.044
(iii) Updating DCRt for Balloon and ARM Payment Shocks
The joint probability variable, JPt, is given additional
weight for balloon loans in the maturity year. Weaker loans will be
unable to qualify for refinancing in the balloon year, especially if
there is an increase in rates, which leads to more defaults at that
point, for any given level of DCRt and LTVt. This
effect should be a function of JPt. Balloon year
[[Page 18202]]
shock is added using a composite variable BJPt:
[GRAPHIC] [TIFF OMITTED] TP13AP99.045
where BYRt is a dummy variable equal to 1 if the observation
is the balloon year, and 0 otherwise, and JPt is the joint
probability of negative equity and negative cash flow. (The loan
specific subscript, j, is dropped here for ease of exposition.) Due to
the small number of balloon loans in negotiated purchase portfolios,
this variable is only estimated in the default rate equation for cash
purchase loans. In stress test application, the estimated coefficient
for cash purchases is also used to predict default rates of negotiated
purchase balloon loans in the maturity year.
The Enterprises tend to extend balloon loans beyond maturity when
properties cannot meet minimum qualification standards for a new loan,
provided the borrower continues to make the monthly payment on the
original mortgage. This possibility of what is called ``extension
risk,'' the risk of loans not leaving the portfolio at the balloon
point, has been documented by Elmer and Haidorfer (1997) and by Abraham
and Theobald (1997). OFHEO also finds that in the Enterprise database a
large percentage of loans are extended beyond balloon maturity. This
model imposes payment shock for extended loans by updating the DCR to
reflect what the borrower would be paying if the borrower refinanced
the property. DCRt is updated after the balloon point by
adjusting PMTt to reflect a new payment level commensurate
with market interest rates for fixed-rate (fully amortizing) loans in
the balloon year.
ARMs are treated with similar DCR adjustments, except that the
payment adjustment occurs annually.\267\ Fannie Mae and Freddie Mac
purchased very few ARM loans through their cash programs, however there
are significant numbers of negotiated transactions that are ARMs.
---------------------------------------------------------------------------
\267\ Nearly all ARMs in Enterprise portfolios are indexed to
the 11th District FHLB Cost of Funds, with monthly rate adjustments,
semi-annual payment adjustments, and negative amortization
provisions. The payment adjustment calculations here proxy for the
full stress of partial payment adjustments and negative amortization
by treating ARM loans as 5/1 products where annual payment changes
are only limited by the lifetime and annual rate caps (5 and 1
percent, respectively). This allows for larger potential payment
shock than would normally be allowed on these loans to compensate
for the lack of negative amortization provisions in this model.
---------------------------------------------------------------------------
(iv) The Present Value of Depreciation Write-offs for Multifamily
Properties
The value of depreciation write-offs to a new property owner is
calculated with the present value formula used by Goldberg and Capone
(1998): \268\
---------------------------------------------------------------------------
\268\ The variable in the Goldberg and Capone (1998) article is
called PVTAX, but it is the same as the DW variable shown here.
Weights for and are the percent of taxpayers
in adjusted gross income groups.
[GRAPHIC] [TIFF OMITTED] TP13AP99.046
DWt is the present value of depreciation write-offs for
each $100 of investment in rental housing, and can be thought of as the
percentage of the investment tax basis that is returned to the investor
through depreciation write-offs. The tax rate data used to calculate
this variable are described above in section IV.D.2., Historical Data.
In addition to tax rates, an estimate of a required rate of return
is needed to calculate the present value of depreciation write-offs.
For this OFHEO used an estimate of the weighted average cost of
capital, with 20 percent equity and 80 percent debt financing. The cost
of debt financing is measured with data from the Enterprises on the
average coupon rate of multifamily fixed-rate mortgages in each year,
1983-95 (rf,t). The cost of equity is calculated with data
from the Enterprises and the Bureau of Labor Statistics. In particular,
if property NOI is expected to increase annually at the rate g, then
the cap rate, CAP, can be thought of as equaling the required return on
equity (re) minus the growth rate, gt. This
implies that the required return on equity equals:
[GRAPHIC] [TIFF OMITTED] TP13AP99.047
CAP0,t is estimated using cap rate values for all
Enterprise loans originated in year, t, and the relationships estimated
in equation 27. Values for gt are three-year average growth
rates of rents, using the Bureau of Labor Statistics CPI residential
rent series, national average (for years t-2, t-1, and t).
The weighted average discount rate for all loans in year, t, is
then:
[[Page 18203]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.048
Table 34 shows values of DWt in the study period, 1983-
95.
[GRAPHIC] [TIFF OMITTED] TP13AP99.222
(v) Program Restructuring
The original cash purchase programs of the Enterprises were
implemented in an overheated lending environment in which appraisal
practices allowed for inflation adjustments to rents when calculating
property value. Such adjustments resulted in understatements of
LTV0 and overstatements of DCR0, leading to the
purchase of loans with understated credit risk and, eventually, to
severe credit losses. In addition to the overstatement of anticipated
rents, original multifamily cash-purchase programs at the Enterprises
had other significant weaknesses. For these reasons, on loans purchased
under original cash programs (Fannie Mae, 1983-1987, Freddie Mac, 1983-
1991) the stress test accounts for increased risk in two ways. The
first method is to adjust LTV0 and DCR0 on
original cash program loans to extract the average inflation factors.
Internal research at OFHEO has concluded that reasonable adjustment
multipliers are 0.85 for DCR0 and 1.27 for LTV0.
The second method used to account for increased default risk in
original cash programs is to include a dummy variable (PR) in the
default equation. This measures the behavioral difference of loans
purchased prior to program restructuring (1 = original cash purchase
loan).
(vi) Default Type
For most loans acquired through negotiated transactions, the loan
event used to estimate defaults is a 90-day delinquency, rather than a
foreclosure. A different event was chosen for these loans because the
seller/servicer typically has a contractual obligation to repurchase
delinquent loans from security pools and resolve the default. As a
result, the Enterprises' data do not reflect which of these loans were
cured or renegotiated and which resulted in property loss events. These
loans will have more observed ``defaults'' because they include cures
and loan modifications as well as property loss events. To adjust for
this discrepancy, two dummy variables are included in the negotiated
purchase default equation: one to flag ARM loans under repurchase
contracts (RA), and one to flag fixed-rate loans under repurchase
contracts (RF).
(vii) Loan Age
Default risk is greatest in the years just after loan origination.
Apartment projects are then most vulnerable to economic shocks because
DCRt may be low, LTVt may be high, and it may
take several years to create a viable market niche for the property.
However, a financially troubled project will not default immediately.
First, valuable depreciation write-offs may be available in the early
years to counterbalance negative property cash flow. Second, working-
capital reserves may forestall default. And third, the owner may
``bleed the project'' by deferring maintenance and other expenditures
prior to delinquency.\269\ Age denotes the loan year of an observation.
Thus, if a loan was originated in 1985, its age is 1 in 1985, 2 in
1986, and so on.
---------------------------------------------------------------------------
\269\ This final reason is discussed by Quercia (1995) and by
Riddiough and Thompson (1993).
---------------------------------------------------------------------------
Other studies of commercial mortgage defaults confirm that defaults
tend to rise in the first years after loan origination and then, once
the weakest loans exit, the conditional default rate declines.\270\
Preliminary analysis of Enterprise data indicated that the peak
[[Page 18204]]
default period is about four years after loan origination. To capture
this underlying trend, a quadratic age function is included in the
default equations.
---------------------------------------------------------------------------
\270\ See Snyderman (1994).
---------------------------------------------------------------------------
b. Prepayment Equations
The explanatory variables chosen for the prepayment equations are
designed to capture multiple refinancing incentives: exercising the
``call'' option (normal refinance); rebalancing debt and equity in the
property (cash-out refinance); risk aversity with respect to pending
balloon expirations (early payoffs); and balloon payoffs. The overall
model is separated into five equations in order to best capture the
differing prepayment incentives by product and product-life stage. For
ease of exposition, these five equations are referred to here as
``models.''
The first model is for fixed-rate loans in the initial yield
maintenance period, when refinancing has no immediate value. Beyond the
yield maintenance period, fully amortizing and balloon loans with fixed
interest rates are analyzed separately in two additional models. This
approach is used because, after yield maintenance ends, balloon loans
prepay more quickly than self-amortizing loans, reflecting borrower
uncertainty surrounding interest rate movements leading up to the time
of loan maturity, when a payoff is required. At maturity, balloon loans
are viewed as having payoffs rather than prepayments. The dynamics of
required payoffs are much different from those of voluntary prepayments
prior to maturity. Therefore, a fourth equation is estimated for
balloons during and after the maturity year. This fourth model includes
both fixed-and adjustable-rate balloons. The fifth and final model is
for adjustable-rate mortgages other than those that may have reached a
balloon maturity point. Adjustable rate mortgages do not have yield
maintenance terms, and their refinancing incentives are different from
those of fixed-rate mortgages.
In prepayment model 4, for balloon payoffs, OFHEO recognizes that
while there is a contractual obligation to find new sources of
financing at the balloon point, those with weak financials may not
qualify for new funding. The Enterprises, like all lenders, however,
are often unwilling to initiate foreclosure if loan payments are being
made under the current (but now expired) contract. OFHEO's approach to
these extended loans is, therefore, to continue to model payoff rates
at and beyond the balloon point.
Table 35 sets forth the structure of the explanatory variables used
in the five prepayment equation/models, as follows:
[[Page 18205]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.223
(i) Relative Spreads in Interest Rates
The relative difference between coupon and market interest rates is
the primary call option variable used in the prepayment equations. For
fixed-rate loans (prepayment models 1-3), OFHEO includes spread
variables when market rates are lower (RSDj,t) and when
market rates are higher than coupon rates
[[Page 18206]]
(RSUj,t). Asymmetry of effects is allowed for because drops
in rates affect refinancings with different motivations than rises in
rates do. Rate declines stimulate refinancings designed to lower
interest costs, while rate increases discourage cash-out refinancings.
[GRAPHIC] [TIFF OMITTED] TP13AP99.049
[GRAPHIC] [TIFF OMITTED] TP13AP99.050
[GRAPHIC] [TIFF OMITTED] TP13AP99.238
The down-rate spread variable, RSDj,t, is given added
weight in the years preceding balloon maturity (model 2) in order to
capture the risk aversity of borrowers with respect to interest rate
movements leading up the balloon point. This weight is added through
two interactive variables. First, RSD1j,t, is
RSDj,t multiplied by a 0/1 dummy variable that is turned on
during the year immediately preceding the balloon year (13-24 months
prior to the maturity month). The second, RSD2j,t, is
RSDj,t multiplied by a 0/1 dummy variable that is turned on
during the second year preceding the balloon year (months 25-36 prior
to the maturity month).
For adjustable rate mortgages (model 5), the spread variable is not
separated into positive and negative components, but is allowed to have
one effect for both increases and decreases in interest rates.\271\
Because ARM coupon rates change every year, the relative spread
variable is used to capture the slope of the yield curve, which
indicates whether it is more valuable to retain the ARM or to refinance
into a fixed-rate loan.
---------------------------------------------------------------------------
\271\ Also, a lack of observations on high interest rate
environments made it difficult to estimate separate effects for rate
rises (RSU).
[GRAPHIC] [TIFF OMITTED] TP13AP99.051
(ii) Market Interest Rate
An additional interest rate variable is added to the ARM equation
(model 5). This is the fixed-rate mortgage rate, rf,t, and
it captures incentives to refinance into fixed-rate products when the
level of rates is low.
(iii) Years-To-Go in the Yield Maintenance Period
Yield maintenance fees are a function of the remaining time until
the end of the prepayment restriction period. As the yield maintenance
period draws to a close, the prepayment penalties decline and the value
of refinancing increases. To capture this change, prepayment model 1
has a variable that measures the years-to-go until the end of the yield
maintenance period (YTGt).\272\
---------------------------------------------------------------------------
\272\ OFHEO experimented with variables that attempted to
capture the impact of yield maintenance fees on refinancing
incentives, but the fixed effects (years-to-go) proved to be a
better predictor of historical mortgage performance.
---------------------------------------------------------------------------
A small number of older Enterprise loans had prepayment lockouts
for a period of years, rather than financial prepayment fees. For these
loans, we set YTGt equal to 10 (its maximum value)
throughout the restriction period.
(iv) Loan-to-Value Ratio
Investors in multifamily properties will engage in cash-out
refinancings to increase returns on invested equity. This refinance
motivation as LTV falls over time is captured by including
LTVt as an explanatory variable.
(v) Loan Age
The baseline prepayment hazard is a function of the desired holding
period of investors. The holding period is heavily influenced by tax
laws: accelerated writeoffs and shorter depreciation schedules
encourage shorter holding periods. It is also affected by exogenous
factors, e.g., investor retirement. Lacking data to measure the
expected holding periods of investors, we assume that the distribution
of expected holding periods, and their effect on baseline prepayment
rates, can be captured through a quadratic function of mortgage
age.\273\
---------------------------------------------------------------------------
\273\ Follain, et al. (1997) attempt a fourth-order function of
age to provide a more flexible baseline hazard function, but the
third and fourth order terms are not statistically significant.
Therefore, OFHEO accepts a second-order age function as sufficient
for capturing the distribution of expected investor holding periods.
---------------------------------------------------------------------------
[[Page 18207]]
(vi) Probability of Qualifying To Refinance
An important obstacle to call option exercise is qualifying for a
new loan. Because information on property financials after loan
origination is not available, it is not known which properties can, at
any point in time, meet minimum standards, DCR=1.20 and LTV=0.80.
Instead, the model uses the same approach employed for default
analysis, calculating the joint probability that DCR and LTV will meet
minimum qualification standards (PQt). PQt is
measured by evaluating the bivariate normal distribution shown in
equation 42 with new integration limits:
[GRAPHIC] [TIFF OMITTED] TP13AP99.052
where, for any given loan (j) in any given time period (t):
[GRAPHIC] [TIFF OMITTED] TP13AP99.053
[GRAPHIC] [TIFF OMITTED] TP13AP99.054
[GRAPHIC] [TIFF OMITTED] TP13AP99.239
This effectively estimates the probability:
[GRAPHIC] [TIFF OMITTED] TP13AP99.055
(vii) Summary of Prepayment Models
In summary, the five prepayment models (equations) are organized as
follows:
1. Model 1: All Fixed-Rate Mortgages-Fully Amortizing and Balloon-in
the Yield Maintenance Period
Includes explanatory variables to capture investor holding horizons
(AYt, AYt2), normal refinancings
(RSDt), cash out refinancings (LTVt), adverse
interest rate effects on cash-out refinancings (RSUt), and
effects on normal refinancings due to yield maintenance
(YTGt).
2. Model 2: Balloon Loans After Yield Maintenance, but Prior to the
Maturity Year
Includes explanatory variables for normal refinancings
(RSDt), cash-out refinancings (LTVt), preballoon
incentives to refinance and avoid the uncertainty of interest rates at
maturity (RSD1t and RSD2t), and the various
investment horizons of borrower/owners (AYt,
AYt2). The variable for adverse interest rate
offsets to cash-out refinancings (RSUt) is not included in
this equation because of a lack of positive observations in the
historical data series.\274\ The coefficient from model 3 is used for
this variable in this equation in stress test application.
---------------------------------------------------------------------------
\274\ Estimating the regression equation with both RSD and RSU
does not significantly change the coefficient on RSD. The RSU
coefficient is negligible and without statistical significance.
---------------------------------------------------------------------------
3. Model 3: Self-Amortizing Fixed-Rate Loans After Yield Maintenance
Includes explanatory variables for investment horizons
(AYt, AYt2), normal refinancings
(RSDt), cash-out refinancings (LTVt), and adverse
interest-rate effects on cash-out refinancings (RSUt).
4. Model 4: Balloon payoff
Includes an explanatory variable for the ability of the property to
qualify for new financing (PQt). This is the only variable
because at the balloon point there are no longer prepayments, only
payoffs.
5. Model 5: Prepayments of Adjustable Rate Mortgages
Includes explanatory variables for the expected investment horizons
of borrower/owners (AYt, AYt2), cash-
out refinance incentives (LTVt), and incentives to refinance
out of ARMs and into fixed-rate products (RSt and
rf,t).
5. Results of the Statistical Estimation of Default and Prepayment
Equations
Table 36 provides maximum likelihood estimates of coefficients in
the two default equations.
[[Page 18208]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.224
All coefficient signs are as expected in both default equations,
and all variables have significant effects, both statistically and
practically. The age patterns in each equation (including the constant
term) are similar, but the joint probability (JPt) has a
larger effect on negotiated purchase default rates than on cash-
purchase default rates. This finding may result from the fact that most
negotiated loan defaults were 90-day delinquency rather than
foreclosures, and delinquencies may be more sensitive to changes in
variables, such as vacancy rates, that underlie JPt.
The dummy variable for program restructuring (PR) has a coefficient
of 0.6203. That implies that annual default rates on original cash-
purchase loans are roughly 1.6 times those of new-cash purchase
loans.\275\ The value of the depreciation write-off coefficient
indicates that the decrease in depreciation allowances that were part
of the 1986 tax reform increased default rates roughly 40 percent.\276\
---------------------------------------------------------------------------
\275\ The marginal probability of binary logit coefficients is
P(1-P), where is the coefficient and P is
the probability estimated with the coefficient set to zero. So, if
P=1 percent, then the increase in probability for original cash
program loans is equal to 0.61 percent, and the original-program
probability is 1.61 percent. If P=0.5 percent, then the probability
for an original-program cash loan is 0.8 percent (marginal
probability is 0.30 percent).
\276\ This finding is explored in greater depth in Goldberg and
Capone (1998).
---------------------------------------------------------------------------
Table 37 provides maximum likelihood estimates of the five
prepayment models (equations). All of the coefficient estimates have
the expected signs and provide consistent results. While the
coefficient of the negative spread variable (RSDt) is larger
during the yield maintenance than it is out of yield maintenance, it
actually has a much smaller effect on the probability of prepayment. In
this functional form, the coefficient represents (approximately) the
percentage change in prepayments per unit change in rates. Because
prepayment rates are much greater for loans out of yield maintenance,
the larger proportional effect for loans in yield maintenance is still
much smaller in absolute terms.
[[Page 18209]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.225
As expected, balloon loans in the post-yield maintenance period
have higher refinance incentives than do fully-amortizing loans, and,
therefore, there is a higher coefficient on RSDt in
prepayment (model 2) than in prepayment (model 3), with even greater
effects as balloon maturity approaches (RSD1t and
RSD2t).
Cash-out refinancings (LTVt), are much stronger in the
post-yield maintenance period, than during yield maintenance, as
expected. ARM loan prepayments (model 5) are sensitive to all of the
factors in the model. The balloon payoff (model 4) shows that the
probability of qualifying for a refinancing is a valuable predictor of
annual payoff rates in the balloon and post-balloon years.
[[Page 18210]]
6. Application to the Stress Test
The risk-based-capital stress test matches default and prepayment
models for each loan group by loan characteristics and age. Because the
stress test uses loan aggregates (groups), the probabilities that
result from use of the statistical equations can be thought of as rates
of default and prepayment on the outstanding balances in each loan
group, in each month of the stress period. But the default and
prepayment models generated here produce annual rates of default and
prepayment. Monthly rates are derived by first calculating annual
equivalent rates in each month, given the explanatory variable values
in that month, and then converting the annual rates to their monthly
equivalents.
The stress tests selects the appropriate default equation used for
each loan group based solely on the value of the Program Type data
field in the Enterprises' loan characteristics data. The stress test
chooses among prepayment equations based upon Product Type and loan
Origination Term fields in the loan characteristics data, and also upon
a computed mortgage age variable. Balloon loans use three separate
prepayment models throughout loan life: in yield maintenance (model 1),
post-yield maintenance (model 2), and payoff period (model 4). Fully
amortizing ARMs will use just one equation (model 5), balloon ARMs will
use two equations (model 5, and then model 4 at balloon term). Fully
amortizing fixed-rate loans will use two prepayment equations, model 1
during yield maintenance and model 3 afterward. The estimated default
and prepayment equations are used not in binary logistic equations, but
rather in trinomial equations, as shown in equations 20 and 21, above.
Use of the trinomial or, more generally, multinomial probability
equations assures that prepayments and defaults are treated as
competing risks in stress test application.
Use of the statistical equations in the stress test also involves
some cross-equation grafting of coefficients. This is because the
historical data on post-yield maintenance balloon loans (model 2) do
not have sufficient observations where market interest rates are higher
than coupon rates to compute a reliable coefficient for RSU. Instead,
the coefficient for the variable RSUt in model 3 is added
into model 2 so that the effect of the up-rate stress test can be
captured. An additional cross-equation grafting is performed for the
added balloon-year effect for the joint probability variable in the
default equations (BJP). There are insufficient loans with balloon
maturity in the negotiated purchase data set to estimate a coefficient.
Therefore, the coefficient estimate from the cash purchase default
equation is used in the negotiated purchase default equation in the
stress test.
The cap rate multiplier used to update property value from NOI
(equation 27) is updated in the stress test using ten-year constant
maturity Treasury yields, rather than mortgage coupon rates. Which
interest rate is used to capture percent changes in interest rates is
not important, and the ten-year constant maturity Treasury Yield series
is the fundamental interest rate series used in the stress test. The
stress test also uses a simplifying assumption for the depreciation
writeoff variable, DWjt. Rather than predict the value of
this variable into the future, OFHEO chose to use the 1995 value (9.27)
for the entire stress period, in both up- and down-rate scenarios.
7. References
Abraham, Jesse (1993a). A Cash Flow Model of Property Performance,
unpublished manuscript, Freddie Mac Corporation, McLean, Virginia,
December.
Abraham, Jesse (1993b). Credit Risk in Commercial Real Estate Lending,
unpublished manuscript, Freddie Mac Corporation, McLean, Virginia,
December.
Abraham, Jesse M. and H. Scott Theobald. 1997. ``Commercial Mortgage
Prepayments,'' in Frank J. Fabozzi and David P. Jacob, eds., The
Handbook of Commercial Mortgage-Backed Securities. New Hope, PA: Frank
J. Fabozzi Associates, 55-74.
Altman, Edward I. (1983). Corporate Financial Distress. New York: John
Wiley & Sons, 1983.
Barnes, Walter C. and S. Michael Gilberto (1994). A Model for Assessing
Commercial Mortgage Risk-Based Capital Factors, mimeo.
Begg, C.B. and R. Gray (1984). ``Calculation of Polychotomous Logistic
Regression Parameters Using Individualized Regressions,'' Biometrica,
71:11-18.
Boyer, Lawrence G., James R. Follain, Jan Ondrich, and Richard A.
Piccirillo, Jr. 1997. A Hazard Model of Prepayment and Claim Rates for
FHA Insured Multifamily Mortgages, unpublished paper. Arlington, VA:
Price Waterhouse, December.
Brennan, M. and E. Schwartz (1985). Evaluation Natural Resource
Investments, Journal of Business, 135-157.
Capone, Jr., Charles A. (1991). Bankruptcy, Survival, and the
Multifamily Mortgage: A Methodological Primer for HUD Staff,
unpublished manuscript, Washington, DC, U.S. Department of Housing and
Urban Development, August.
Childs, Paul D., Steven H. Ott, and Timothy J. Riddiough. 1995. The
Pricing of Multi-Class Commercial Mortgage-Backed Securities.
Unpublished manuscript, MIT Center for Real Estate, February.
Danter, Kenneth F. (1996). ``Smart Investments in Multifamily,''
Mortgage Banking, July, 74-79.
DiPasquale, Denise and Jean L. Cummings (1992). ``Financing Multifamily
Rental Housing: The Changing Role of Lenders and Investors,'' Housing
Policy Debate 3 (1), 77-116.
Dyl, E.A. and H.W. Long (1969). Abandonment Value and Capital
Budgeting: Comment, Journal of Finanace, 88-95.
Elmer, Peter J. and Anton E. Haidorfer. 1997. ``Prepayments of
Multifamily Mortgage-Backed Securities.'' The Journal of Fixed Income
(March), 50-62.
Follain, James R. and Jan Ondrich. 1997. ``Ruthless Prepayment?
Evidence from Multifamily Mortgages,'' Journal of Urban Economics 41
(January), 78-101.
Follain, James R., Jan Ondrich and Sinha P. Gyan (1990). A Hazard Model
of Multifamily Mortgage Prepayments. Discussion Paper No. 54,
Department of Economics, Syracuse University, December 1990.
Foster, Chet and Robert Van Order (1984). ``An Option Based Model of
Mortgage Default,'' Housing Finance Review 3 (4), 351-372.
Goldberg, Lawrence (1994). Claims Forecasting Models for FHA
Multifamily Housing Loans. Reston, Virginia: Economic Research
Laboratory, Inc., under contract to the U.S. Department of the Treasury
through HCI Inc., Reston, Virginia, August 26.
Goldberg, Lawrence and Charles A. Capone, Jr. 1997. Motivation and
Testing of a Double-Trigger Hypothesis for Multifamily Loan Defaults.
Unpublished manuscript, Office of Federal Housing Enterprise Oversight:
Washington DC.
Goldberg, Larry, and Capone, Jr., Charles A. (1998). ``Multifamily
Mortgage Credit Risk: Lessons from Recent History,'' Cityscape 4 (1),
93-113.
ICF Incorporated (1991). Predicting Financial Distress In HUD
Multifamily Projects. Fairfax, Virginia: ICF
[[Page 18211]]
Incorporated, under contract with the U.S. Department of Housing and
Urban Development, April 30.
Joy, O.M. (1976). Abandonment Values and Abandonment Decisions: A
Clarification, Journal of Finanace, 1225-1228.
Kahn, Charles M. (1991). The Economic Role of Foreclosure Rules, ORER
Letter, Spring, 8-11.
Kau, James B., D.C. Keenan, W.J. Muller III and J.F. Epperson (1987).
The Valuation and Securitization of Commercial and Multifamily
Mortgages, Journal of Banking and Finance 11, 525-546.
Kau, James B., D.C. Keenan, W.J. Muller III and J.F. Epperson (1990).
``Pricing Commercial Mortgages and Their Mortgage-Backed Securities,''
Journal of Real Estate Finance and Economics 3, 333-356.
Mahue, Michelle A. (1991). ``The Economic Role of Statutory
Redemption,'' ORER Letter, Spring, 12-13.
McFadden, Daniel (1975). ``The Revealed Preferences of a Government
Bureaucracy: Theory.'' Bell Journal (Autumn), 401-416.
National Task Force on Financing Affordable Housing (1992). From the
Neighborhoods to the Capital Markets. Washington, DC: Allstate
Insurance Company, June.
Pedone, Carla I. (1991). ``Estimating Mortgage Prepayments and Defaults
in Older Federally Assisted Rental Housing and the Possible Costs of
Preventing Them,'' Housing Policy Debate 2 (2), 245-288.
Quercia, Roberto (1995). On Developing a Model of Mortgage Default for
Multifamily Rental Housing, unpublished manuscript. Washington, DC: the
Urban Institute, October.
Riddiough, Timothy J. and H.E. Thompson (1993). ``Commercial Mortgage
Pricing with Unobservable Borrower Default Costs,'' Journal of the
American Real Estate and Urban Economics Association 21 (3), 265-292.
Riddiough, Timothy J. and S.B. Wyatt (1994a). ``Strategic Default,
Workout and Commercial Mortgage Valuation,'' Journal of Real Estate
Finance and Economics 9 (1), 5-22Riddiough, Timothy J. and S.B. Wyatt
(1994b). ``Wimp or Tough Guy: Sequential Default Risk and Signaling
with Mortgages,'' Journal of Real Estate Finance and Economics 9 (3),
299-322.
Robicheck, A. and J.C. Van Horne (1967). Abandonment Value and Capital
Budgeting, Journal of Finance, 577-590.
Sharkawy, M. Atef and Walter C. Barnes (1992). ``Cost, Value, and
Hybrid-Based Underwriting Criteria,'' The Journal of Real Estate
Research 7 (2, Spring), 169-185.
Snyderman, Mark P. (1994). ``Update on Commercial Mortgage Defaults,''
The Real Estate Finance Journal, Summer, 22-32.
Titman, S. and W. Torous (1989). ``Valuing Commercial Mortgages: An
Empirical Investigation of the Contingent Claims Approach to Pricing
Risky Debt,'' Journal of Finance 44, 345-373.
Standard & Poors (1993). Credit Week, March 8. (Issue devoted to
Commercial Mortgage Securities)Vandell, Kerry (1992). ``Predicting
Commercial Mortgage Foreclosure Experience,'' Journal of the Americal
Real Estate and Urban Economics Association 20 (1), 55-88.
Vandell, Kerry, Walter Barnes, David Hartzell, Dennis Kraft and William
Wendt (1993). ``Commercial Mortgage Defaults: Proportional Hazards
Estimation Using Individual Loan Histories,'' Journal of the American
Real Estate and Urban Economics Association 4 (21, Winter), 451-480.
E. Multifamily Loss Severity
1. Introduction
Owing primarily to limited available data, OFHEO's approach to
modeling multifamily loss severity rates for stress test application is
simpler than approaches chosen for other elements of mortgage
performance. The number of multifamily loans in Enterprise portfolios
is a fraction of the number of single family loans. Therefore, the
number of defaulted multifamily loans is relatively small. Further,
only one Enterprise, Freddie Mac, has reliable historical records of
multifamily loss severity rates. Until the mid-1990s, Fannie Mae's
multifamily default resolutions were handled by the various field
offices, and there were no standard protocols for tracking and
maintaining data elements on a loan-by-loan basis. The result is that
OFHEO analysis of Enterprise experience is exclusively focused on that
of Freddie Mac.
Even so, the Freddie Mac program provides sufficient data to
understand the various components of loss severity rates. They
represent the worst historical experience of the Enterprises, and some
of the worst experience on record for industry-wide multifamily
mortgage loss severities. The Freddie Mac data are not extensive enough
to allow a multivariate statistical analysis. The analysis outlined
here is univariate: each element is examined individually, without
explanatory variables. The result is that OFHEO chose for its stress
test to use simple averages of various components of multifamily loss
severity.
Section 2 of this supplementary material on multifamily loss
severity gives an outline of the conceptual framework, the plan OFHEO
used in approaching multifamily loss severity rates; section 3 provides
a discussion of the source data; section 4 is a summary of the data
analysis; and section 5 concludes with an examination of how the loss
severity components are applied in the stress test.
2. Conceptual Framework
Loss severity is the net cost of resolving a mortgage default. It
is most typically measured as a percentage of the unpaid principal
balance (UPB) at the time of default.\277\ OFHEO measures severity in
this way and then applies any available credit enhancements against the
loss to arrive at a net loss to the Enterprises. Credit enhancements
are not discussed in this supplement. A description of how the stress
test applies credit enhancements can be found in the Appendix to this
regulation.
---------------------------------------------------------------------------
\277\ All references to UPB in this part of the supplement
indicate UPB at time of default.
---------------------------------------------------------------------------
OFHEO's general approach is to model only those loss severity rates
associated with full foreclosure events. The one exception is for
programs where the default event of record is a 90-day delinquency.
This exception will be discussed below, under Data Analysis.
Foreclosure results in the Enterprise taking title to the property,
managing and rehabilitating it, and then marketing and selling the
property. OFHEO also models the timing of events and cost elements
associated with foreclosure and property management. As with single
family loss severity rates, OFHEO recognizes three time frames in
capturing costs and revenues associated with mortgage foreclosure: the
first four months of delinquency, the time from default to foreclosure
completion (which includes the first four months), and time of property
inventory (from foreclosure completion to property disposition).
After analyzing Enterprise data, and reviewing available research
on multifamily loss severity, OFHEO chose to use simple averages of
Enterprise experience, by loss component, and not to perform
multivariate statistical analysis. Component analysis permits the use
of discounting techniques to create effective loss severity rates at
the time of default (one month after last-
[[Page 18212]]
paid-installment). OFHEO found no basis in the existing literature for
multivariate statistical analysis of multifamily loss severities.
OFHEO identified seven studies of loss severity, each of which
relies upon data from a broad range of commercial property types, and
each of which defines and measures severity rates somewhat
differently.\278\ These studies primarily provide simple averages of
loan-level loss severity rates, though some do attempt some statistical
analysis of severity rates. Curry, et al (1990) model loss severities
as a function of the type of organization managing the foreclosed
property (public or private). Haidorfer (1997) performs a multivariate
statistical analysis that looks at the type of property sale process
(open auction, sealed-bid auction, or broker sales). He finds that the
type of selling process does not influence severity rates. A third
study by Ciochetti and Riddiough (1998) models expected property
recovery rates as a function of mortgage terms, and a list of property
type and region dummies.\279\ They find no statistical significance of
original LTV, debt coverage ratio (DCR), loan age, or the mortgage
interest rate.
---------------------------------------------------------------------------
\278\ The seven studies are: Curry, Blalock, and Cole (1990);
Snyderman (1994); Fitch Investors Service (1996); Ciochetti (1997);
Haidorfer (1997); Barnes, Gilberto and Peyton (1998); and Ciochetti
and Riddiough (1998).
\279\ The Ciochetti and Riddiough study looks at expected
recoveries immediately following foreclosure, where property value
is appraised value, and no property management or disposition costs
are included in the calculations.
---------------------------------------------------------------------------
3. Sources of Data
OFHEO obtained loss severity data on multifamily loans from both
Enterprises, but only Freddie Mac maintained a complete historical data
base of all relevant revenue and expense components that was useful for
this analysis.\280\ The analysis of foreclosure loss severities is then
limited to 705 multifamily loans purchased by Freddie Mac, that
subsequently defaulted between 1987 and 1995 and ended in foreclosure.
Over 83 percent of these loans defaulted between 1990 and 1993, in what
is considered the worst period in modern history for the commercial
mortgage market. These data are supplemented by Freddie Mac data on
other default resolutions. These additional data are used for
projecting potential losses on negotiated purchase loans for which
seller/servicers must repurchase and resolve all 90-day delinquencies.
Once delinquencies are resolved, the seller/servicers bill the
Enterprise for the net costs.\281\ Fannie Mae has a large portfolio of
sold loans with these repurchase provisions and has maintained data on
the claims for losses submitted by the seller/servicers. However, many
of the claim records are incomplete and OFHEO therefore, relied on
information on Freddie Mac default resolutions, and on information from
other available studies, to determine a loss rate to charge against 90-
day delinquencies. Freddie Mac provided OFHEO with information on the
chargeoffs associated with 160 non-foreclosure resolutions that
occurred from 1990 to 1995.
---------------------------------------------------------------------------
\280\ Until the mid 1990s, Fannie Mae's foreclosed property
inventory was managed by the individual field offices. There were no
standard protocols for recording or retaining expense and revenue
components of loss severity on a loan-by-loan basis. Fannie Mae
could only provide OFHEO with consistent data on event times
(foreclosure and property disposition).
\281\ When these loans are purchased by the Enterprises, the
seller/servicers must establish resource accounts. These credit
enhancements drawn on as first-lost protection before the
Enterprises actually incur any costs from loan defaults in these
mortgage pools.
---------------------------------------------------------------------------
These data represent the worst historical experience of the
Enterprises, which began purchasing conventional multifamily mortgages
in 1983.\282\ The Freddie Mac data is among the largest and richest
sets of information available to any researchers who have studied
multifamily loss severities.
---------------------------------------------------------------------------
\282\ Goldberg and Capone (1997) detail the problems that led to
high default rates among multifamily mortgages in the late 1980s and
early 1990s. These same factors led to high severity rates. In
addition to market factors, Freddie Mac attributes its particularly
bad performance to fraud by lenders that underwrote loans that were
not of investment quality. An analysis of data shown in Investor
Analyst Reports shows that in 1991, Freddie Mac's chargeoff for bad
multifamily loans was more than its total chargeoff for bad single
family loans, even though its multifamily portfolio of $10 billion
was only three percent as large as the single family portfolio. This
high rate of chargeoffs lasted from 1989 through 1992.
---------------------------------------------------------------------------
4. Data Analysis
a. Foreclosure Severity Rates
Table 38 provides average values for loss severity components in
the Freddie Mac foreclosure database. The cost components are each
measured as a percent of the UPB at the time of default. These average
rates are also computed using UPB as a weighting factor on each loan.
This weighting provides a more accurate measure of portfolio severity
rates than would a simple average.\283\ The operating loss per month is
the difference between monthly property income (rents) and expenses,
where expenses include property repairs. It is not surprising that this
element is a net cost rather than a net revenue because defaulting
properties will have high vacancy rates and significant needs for
repairs. The net proceeds of property sale is arrived at by subtracting
selling expenses and other prorated expenses (taxes and rents) due at
settlement from the actual sales price of each property. The two time
dimensions reported here are important for discounting the associated
cash flows to arrive at an effective loss severity rate at time of
default (one month after last-paid-installment). One cost element not
shown in Table 38 is the interest passthroughs to security holders
during the initial months of delinquency. In general, loans are
repurchased from security pools by the 120th day of delinquency, so
that four months of passthrough interest must be added to severity
calculations in stress test application.
---------------------------------------------------------------------------
\283\ UPB weighting is also used in the OFHEO single family loss
severity analysis.
[[Page 18213]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.226
Adding the cost components here produces a 54 percent loss
severity. This sum is comparable to what is reported by Fitch (1996) in
its study of commercial mortgage foreclosures. Fitch reports a 56
percent average loss severity rate on foreclosures.\284\ The Fitch
study had an (undefined) interest passthrough component. If added to
the Freddie Mac severity components, a four-month passthrough at eight
percent interest would increase their sum from 54 percent to roughly 58
percent.
---------------------------------------------------------------------------
\284\ It is not clear exactly how many foreclosures there are in
the Fitch data set. Fitch reports 547 costly default resolutions of
60-day delinquencies, of which it appears from other data given in
the report (loss severity rates on foreclosure and non-foreclosure
resolutions) that 147 are foreclosure events.
---------------------------------------------------------------------------
b. 90-Day Delinquency Severities
Deriving a loss rate to use for 90-day delinquency events involves
making inferences on the rate of foreclosure and other costly
resolutions versus non-costly resolutions. Snyderman (1994) found that
46 percent of 90-day delinquencies in life insurance company
portfolios, 1972-1986, ended in foreclosure. Freddie Mac data are
consistent with this finding. Freddie Mac data indicate that
foreclosures plus other costly resolutions are 56 percent of total 90-
day delinquencies. Using 56 percent as the rate of costly loan
resolutions, and applying a 70 percent foreclosure loss severity to
them, produces a severity rate on 90-day delinquencies of just over 39
percent.\285\
---------------------------------------------------------------------------
\285\ The 70 percent loss rate on foreclosures comes from the 54
to 58 percent reported earlier, with asset holding costs added.
---------------------------------------------------------------------------
5. Application to the Stress Test
The loss severity components just described enter the stress test
as cash flows at various points in the default time frame. These cash
flows are discounted by a cost-of-debt interest rate to produce a net-
present-value loss severity rate in the month of default. The use of
discounting provides an implicit funding cost. It reduces the value of
final proceeds by an amount equal to the cost of funding the non-
performing assets (first the loan, and then the property), and it
reduces the value of various expenditures to reflect the fact that cash
is not actually expended in the month of loan default but could be
invested at some rate-of-return for a number of additional months.What
discounting does not include is the cost of funding that portion of the
loan balance that is not recovered in the sale of the foreclosed
property. That portion of funding cost is captured elsewhere in the
stress test by ongoing interest expenses on debt that is in excess of
what can be retired by the property sale proceeds.\286\ The ongoing
interest expenses are captured in other parts of the stress test beyond
the loss severity calculations.
---------------------------------------------------------------------------
\286\ For retained loans, the debt supporting the mortgage UPB
will already be on the Enterprise balance sheets at the time of
default. For sold loans, however, asset funding occurs when the
Enterprise buys the defaulting loan out of its security pool.
---------------------------------------------------------------------------
a. Foreclosure Loss Severity Rate Application
The basic loss severity equation for foreclosure costs has five
elements, as shown in this equation:
[GRAPHIC] [TIFF OMITTED] TP13AP99.056
[[Page 18214]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.240
The first loss element is the UPB of the defaulted loan. It is set
here equal to `1' or 100 percent. For sold loans, it is discounted for
four months, which represents the timing of repurchasing the loan from
the security pool. For retained loans, the UPB is not discounted
because the economic loss occurs at the time of default. The second
loss element is the passthrough interest expense for four months. This
expense is discounted for two months as an approximation to discounting
each month's pass through individually. This element only appears for
sold loans.
The third element of loss severity is the expense incurred to
obtain a foreclosure judgment on the property. This cost includes all
legal expenses for foreclosure and, when necessary, to release a
bankruptcy stay, and other charges that may be incurred to obtain clean
title to the property (e.g., property taxes due). The fourth element is
the cost of operating and maintaining the foreclosured property while
it is REO. And the fifth element is the net proceeds at final property
disposition.
The formula can be applied very simply. Using the cost elements in
Table 38, along with a discount rate, r d,t = .06, and a
passthrough rate, r p = .08:
[GRAPHIC] [TIFF OMITTED] TP13AP99.057
This reduces to 0.622 for sold loans and 0.615 for retained loans. If
we increase the discount rate to 12 percent, the results change to
0.661 for sold loans and 0.673 for retained loans. If the discount rate
were reduced to three percent, the net present value severity rates
would be 0.598 for sold loans and 0.581 for retained loans.
b. 90-Day Delinquencies
For negotiated purchase loans with seller/servicer repurchase
provisions, the stress test discounts to reflect a time lag between the
initial delinquency and the claim payment. In the stress test, seller/
servicer claims on 90-day delinquencies are settled 12 months after
default. Starting with the 39 percent severity rate for foreclosure
alternatives reported above, and discounting for one year, yields a
rate of around 34 to 37 percent, depending on the actual discount rate.
6. References
Barnes, Walter C., Gilberto, Michael, and Peyton, Martha S. (1997).
Commercial Mortgage Loss Severity: What is it and How Should it be
Measured? unpublished manuscript, Mortgage Analytics, West Hartford,
CT.
Ciochetti, Brian A. (1997). Loss Characteristics of Commercial Mortgage
Foreclosures, Real Estate Finance (Spring), 53-69.
Ciochetti, Brian A. and Riddiough, Timothy J.(1997). Foreclosure Loss
and the Foreclosure Process: An Examination of Commercial Mortgage
Performance, unpublished manuscript, University of North Carolina,
Chapel Hill, Department of Finance.
Curry, Timothy, Blalock, Joseph, and Cole, Rebel (1990). Recoveries on
Distressed Real Estate and the Relative Efficiency of Public Versus
Private Management, AREUEA Journal 19 (4), 495-515.
Fitch Investors Service (1996). Trends in Commercial Mortgage Default
Rates and Loss Severity. New York: Fitch Investors Service, Structured
Finance Special Report, November 11.
Goldberg, Larry, and Capone, Jr., Charles A. (1998). Multifamily
Mortgage Credit Risk: Lessons from Recent History, Cityscape 4 (1), 93-
113.
Haidorfer, Anton E. (1997). Relative Gross Recovery Rates on the Sale
of Distressed Real Estate Owned Evidence from the Resolution Trust
Corporation (RTC), unpublished manuscript, Mortgage Bankers Association
of America, May.
Snyderman, Mark P. (1994). Update on Commercial Mortgage Defaults, The
Real Estate Finance Journal, Summer, 22-32.
F. Property Valuation
1. Introduction
The stress test simulates mortgage performance under housing market
[[Page 18215]]
conditions that reflect stresses comparable to those of the time and
place of the benchmark loss experience (BLE). This section describes
the data used to define and create variables that comprise the housing
market conditions of the stress test.
Three housing market condition variables are used in the stress
test: house price growth rates, rent growth rates, and rental vacancy
rates. House price growth rates are used to project single family
mortgage performance, both default/prepayment rates and loss severity
rates. Rent growth rates and vacancy rates are used to project
multifamily default and prepayment rates.
Section 2 of this part of the Technical Supplement describes the
conceptual framework OFHEO used to determine the housing market
condition variables in the stress test. Section 3 lists the sources of
data used to develop these variables. Section 4 then describes the
statistical analysis performed to transform source data into housing
market condition variables.
2. Conceptual Framework
The BLE is based upon the performance of 30-year, fixed-rate single
family mortgages in four States--Arkansas, Louisiana, Mississippi, and
Oklahoma--originated in 1983 and 1984, during the ten years following
origination, as defined in the first NPR. The actual BLE covered twelve
calendar years because benchmark loans could originate any time between
January 1983 and December 1984, and the ten-year experience of the last
loans originated during the benchmark time period lasted through
December of 1994. For house prices, rent growth rates, and vacancy
rates in the stress test, OFHEO defined the BLE as the years 1984
through 1993--the middle ten years of the twelve-year period marking
the BLE. OFHEO then identified sources of data that reflect the housing
market conditions of the benchmark time and place, and that are
compatible with historical data used to estimate statistical
(econometric) models of mortgage default, prepayment, and loss
severity.
a. Single Family House Price Appreciation Rates
OFHEO sought publicly available data with geographic coverage that
reflect stresses similar to those of the BLE. For house price growth
rates, the stress test uses OFHEO HPI data from the West South Central
(WSC) Census Division. Because the 1984-1993 WSC HPI series was used to
calibrate the single family default- and severity-rate equations to the
actual four-State benchmark loan performance,\287\ the same series also
was used to define housing market conditions in the stress test. The
WSC Census Division is similar geographically to the actual four-State
BLE. The difference is that the WSC includes Texas, but not
Mississippi. For the ten-year period, 1984-1993, the cumulative house
price appreciation rate for the WSC Census Division is very similar to
that of the four-State benchmark region. For the stress test, the OFHEO
HPI is converted from index form into quarterly appreciation rates.
---------------------------------------------------------------------------
\287\ Benchmark loss experience calibration is discussed in both
the Single Family Default/Prepayment and the Single Family Severity
sections of this Technical Supplement.
---------------------------------------------------------------------------
b. Vacancy Rates and Rent Growth Rates
Rental market data--vacancy rates and rent growth rates--used in
the statistical analysis of historical multifamily default and
prepayment rates are also from government sources. Rent growth rates
are from the residential rent component of the consumer price index
(CPI), produced by the Bureau of Labor Statistics. Vacancy rates are
from the rental vacancy rate series (H-111) produced by the Bureau of
the Census. However, these data series are not used directly to reflect
multifamily housing market conditions during the stress period because
the available geographic aggregations and time periods do not closely
match the four-State benchmark. The CPI residential rent index is not
available for the appropriate geographic areas, and the H-111 state
vacancy rate series is not available for 1984 and 1985.\288\
---------------------------------------------------------------------------
\288\ The residential rent series includes MSA level data for
New Orleans, beginning in 1986. The New Orleans data alone, however,
were insufficient for use in representing the BLE.
---------------------------------------------------------------------------
In light of these shortcomings, OFHEO identified a non-government
source of data published by the Institute for Real Estate Management
(IREM). However, the IREM data do not represent the same properties as
the government data. IREM surveys include only apartments, while the
government surveys (both rents and vacancies), include apartments and
single family rental units. To assure consistency with the government
series, statistical regression equations were estimated to use in
adjusting the IREM data. The adjusted data can be thought of as
answering the question, ``What would CPI and H-111 data look like if
they were available in the benchmark area?'' The statistical
regressions (detailed in section 4, Statistical Analysis) use data from
all metropolitan statistical areas (MSAs) for which both IREM and CPI
or H-111 data are available, to estimate statistically valid
relationships. These equations are then applied to IREM data from the
four-State area to assure that variables used in the stress test are
compatible with the variables used to develop the statistical models.
3. Data Sources
The sources of data used to develop the housing market condition
variables for stress test application are as follows:
OFHEO HPI Report, 1996:3, West South Central Census
Division Series, 1983:4-1993:4.
Bureau of Labor Statistics, Consumer Price Index,
Residential Rent Component, MSA series, 1970-1995, annual index values.
Bureau of the Census, H-111 Housing Vacancy Survey, rental
unit vacancies, MSA series, 1981-1995, annual average vacancy rates.
Institute for Real Estate Management. Conventional
Apartments. Chicago, IL: IREM. Annual publications, 1981-1995, MSA
level (median) dollar rents per square foot, (median) dollar vacancy
losses per square foot, and number of apartments in survey.
4. Statistical Analysis
a. House Prices Appreciation Rates
The use of the OFHEO HPI in the stress test requires no statistical
analysis. Monthly house price appreciation rates are derived from the
OFHEO HPI index in three steps. First, monthly appreciation rate
indexes are created for each quarter by dividing that quarter's index
value by the index value for the preceding quarter. Second, the
logarithm of this new index is used as the growth rate factor for that
quarter. Finally, the quarterly rate is divided by three to produce at
monthly growth rate factors for each month in the quarter. In this
manner, the 120 months of stress test HPI growth rate factors
(gq,t) are produced from the 41 quarterly HPI values
(HPIq), 1983:4-1993:4:
[[Page 18216]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.058
[GRAPHIC] [TIFF OMITTED] TP13AP99.241
The gq,t are called growth rate ``factors'' because they
are the continuous growth rate equivalents to actual, discrete growth
as measured across each month and quarter.\289\ Stress test
applications convert these factors to actual appreciation rates. This
baseline series of monthly growth rates applies in both the up- and
down-rate scenarios, but may be adjusted for inflation in the up-rate
scenario.
---------------------------------------------------------------------------
\289\ Continuous growth rates refer to a process whereby house
price appreciation is a continuous process, throughout each month or
quarter. The actual house price index that shows total appreciation
across a month or quarter is just the exponential of the growth rate
factor for that time period.
---------------------------------------------------------------------------
b. Rent Growth Rates
The statistical analysis underlying the rent growth rate variable
used in the stress test uses MSA level data from both IREM and the CPI
for the 26 cities for which the CPI residential rent index is
available.\290\ Annual growth rates for 1970-1995 were computed from
both the IREM and CPI rent data, and the following pooled, time series,
cross-sectional, weighted least squares regression was estimated:
---------------------------------------------------------------------------
\290\ Statistical analysis was based upon what the Bureau of
Labor Statistics calls its ``old series.'' The new series covers 29
MSAs.
[GRAPHIC] [TIFF OMITTED] TP13AP99.059
[GRAPHIC] [TIFF OMITTED] TP13AP99.242
The regression was weighted by the number of apartments that IREM
surveyed in each MSA. The coefficient for IRj,y is
significant at the 99 percent confidence level.
IREM data are available for one city in each of the four benchmark
States--Jackson, Little Rock, New Orleans, and Oklahoma City. A
benchmark region rent growth rate series was computed from equation 57,
using a simple average of annual IREM rent growth rates in each of
these cities (1984-1993) to populate IRj,y. Monthly rent
growth rates were then computed using the following compounding
formula.
[GRAPHIC] [TIFF OMITTED] TP13AP99.060
[GRAPHIC] [TIFF OMITTED] TP13AP99.243
Equation 58 produces final rent growth rates in discrete form,
rather than continuous form, because the process used to create the
original series was discrete. As with the house price growth rate
factors, inflation adjustments may be applied in the up-rate scenario.
c. Vacancy Rates
Because Census vacancy rate data are available at the State level
starting in 1986, OFHEO uses the average of rates
[[Page 18217]]
in the four benchmark States, from 1986-1993 for the latter eight years
of the stress test. For the first two years, OFHEO employs a
statistical analysis similar to that for rent growth rates to create
government-equivalent vacancy rates for 1984 and 1985, the first two
years needed for the stress test. The weighted-least-squares regression
matches MSA-level Census vacancy rates to IREM vacancy rates in the
same cities. Matching data is available for 51 MSAs; 23 with Census
data that begin in 1981, and another 28 for which Census data become
available in 1986. The pooled cross-section, time series regression is:
[GRAPHIC] [TIFF OMITTED] TP13AP99.061
[GRAPHIC] [TIFF OMITTED] TP13AP99.244
The coefficient on IVj,y is statistically significant at
the 99 percent level, but the constant term is not statistically
significant. This lack of significance is not surprising, given that
the regression is relating rates of change and not levels of vacancy
rates. In application, the constant term is dropped from the equation.
To compute vacancy rates for 1984 and 1985, equation 59 is applied
using average IREM vacancy rates for the four benchmark cities to
compute rates of change for the four-State average Census vacancy rate.
The resulting rate of change from 1986 to 1985 is first applied to the
four-State average Census vacancy rate for 1986 to compute a
government-equivalent vacancy rate for 1985. The procedure is repeated
to compute the vacancy value for 1984. Finally, each annual vacancy
rate in the ten-year series is applied to each month in the year to
extend the series to cover the 120 months of the stress period.
V. Regulatory Impact
A. Executive Order 12612, Federalism
Executive Order 12612 requires that Executive departments and
agencies identify regulatory actions that have significant Federalism
implications. ``Federalism implications'' is defined as regulations or
actions that have substantial direct effects on the States, on the
relationship or distribution of power between the national government
and the States, or on the distribution of power and responsibilities
between the Federal and State government. This proposed regulation has
no Federalism implications that warrant the preparation of a Federalism
Assessment in accordance with Executive Order 12612.
B. Executive Order 12866, Regulatory Planning and Review
This regulation has been reviewed by the Office of Management and
Budget (OMB) in accordance with Executive Order 12866. OMB has
determined that this is an economically significant rule. Included in
the preamble to the proposed rule is an economic analysis of the
proposal's impact on the regulated entities, and in particular on
mortgage credit, of various alternatives. It contains a technical
supplement providing detail on the specifications and estimations of
econometric models for mortgage performance, and how those statistical
models are applied in the proposed risk-based capital stress test.
The proposed regulation implements the 1992 Act's requirement that
OFHEO establish a risk-based capital requirement for the Enterprises.
Along with the existing minimum capital leverage ratios and the
examination function, the stress test is designed to ensure that the
Enterprises have adequate capital and operate in a safe and sound
manner.
It is difficult to estimate precisely the particular benefits and
costs associated with the risk-based capital requirement. Where
possible, section II. C., Implications of the Proposed Rule discusses
and quantifies the potential benefits and potential costs in more
detail. Otherwise, that section characterizes the benefits and costs
qualitatively. The analysis indicates that the anticipated benefits
from implementing the risk-based capital regulation outweigh the
anticipated costs. It further indicates that the proposed regulation
ensures that risk is held at an appropriate level, while imposing the
least burden on the Enterprises.
By carrying out Congress' intent to implement the risk-based
capital requirement, OFHEO would reduce the potential for Enterprise
insolvency by protecting against interest rate, credit, and management
and operations risk. By ensuring their safety and soundness, the
regulation allows the Enterprises to continue to carry out their public
purposes.\291\ These include providing stability in the secondary
market for residential mortgages and providing access to mortgage
credit in central cities, rural areas, and underserved areas. In
addition, the regulation will also ensure that the Enterprises will
continue to provide benefits to the primary mortgage market such as
standardizing business practices.\292\
---------------------------------------------------------------------------
\291\ 1992 Act, section 1302(2) (12 U.S.C. 4501(2)).
\292\ Managing Risk in Housing Finance Markets: Perspectives
from the Experiences of the United States of America and Mexico,
OFHEO and the Mortgage Bankers Association of America (June 11,
1998).
---------------------------------------------------------------------------
Other benefits of the risk-based capital requirement are (1) making
the Enterprises' capital requirement more sensitive to differences in
risk exposures, (2) discouraging the Enterprises from taking excessive
risks by making riskier behavior more costly, and (3) ensuring that the
Enterprises maintain adequate capital in stressful credit and interest
rate environments. Implementing a risk-based capital requirement with
credit risk and interest rate risk components will help ensure that the
Enterprises' capital requirement is more closely related to the risks
that they incur. Adopting the proposed rule will result in a capital
requirement that corresponds more closely to capital levels that the
marketplace would demand in the absence of the benefits afforded by the
government sponsorship of the Enterprises, and will lead to gains in
overall economic efficiency.
[[Page 18218]]
Moreover, by evaluating risk in a forward-looking, dynamic manner, the
stress test identifies potential problems before they become
significant.
As detailed in the Implications section, the Proposed Rule may
impose some costs on the Enterprises. Nevertheless, any such costs are
the necessary and reasonable costs of carrying out Congress' intent
that the Enterprises remain financially solvent, which will enable them
to out their important public purposes.
Changes to comply with the risk-based capital requirement can be
accomplished at relatively low costs. Both Enterprises can employ
various practices and procedures to manage credit risk and interest
rate risk by adjusting their holdings or operations. For example, one
method to reduce credit risk exposure is to increase use of credit
enhancements with highly-rated counterparties. One method to reduce
interest risk exposure is to purchase derivative contracts.
By complying with an effective risk-based capital requirement, the
Proposed rule may in fact reduce Enterprise costs by enhancing investor
confidence. This is consistent with a study by Standard & Poor's (S&P)
that provided risk-to-the-government credit ratings for the
Enterprises.\293\ While S&P had rated Fannie Mae A- and Freddie Mac A+
in 1991, the 1997 report upgraded the ratings of both Enterprises to
AA-. S&P cited increased governmental oversight by OFHEO as an
important factor in these higher ratings. It further noted that
``OFHEO's regulatory oversight [of Freddie Mac] also gives comfort that
appropriate interest rate risk mitigation steps would be taken as
needed.'' \294\
---------------------------------------------------------------------------
\293\ Final Report of Standard & Poors to OFHEO, Contract No.
HE09602C (February 3, 1997).
\294\ Contract No. HE09602C, p. 10.
---------------------------------------------------------------------------
C. Executive Order 12988, Civil Justice Reform
Executive Order 12988 sets forth guidelines to promote the just and
efficient resolution of civil claims and to reduce the risk of
litigation to the government. The proposed regulation meets the
applicable standards of sections 3(a) and (b) of Executive Order 12988.
D. Regulatory Flexibility Act
The Regulatory Flexibility Act (5 U.S.C. 601 et seq.) requires that
a proposed regulation that has a significant economic impact on a
substantial number of small entities must include an initial regulatory
flexibility analysis describing the rule's impact on small entities.
Such an analysis need not be undertaken if the agency head certifies
that the rule will not have a significant economic impact on a
substantial number of small entities. 5 U.S.C. 605(b).
OFHEO has considered the impacts of the proposed risk-based capital
regulation under the Regulatory Flexibility Act. The proposed
regulation does not have a significant effect on a substantial number
of small entities.
This proposed regulation would not have a significant economic
impact on a substantial number of small entities since it is applicable
only to the Enterprises, which are not small entities for purposes of
the Regulatory Flexibility Act. Therefore, the General Counsel of OFHEO
acting under delegated authority has certified that the proposed
regulation would not have a significant economic impact on a
substantial number of small entities.
E. Paperwork Reduction Act
The Paperwork Reduction Act of 1995, 44 U.S.C. Chapter 35 requires
that regulations involving the collection of information receive
clearance from the Office of Management and Budget (OMB). The risk-
based capital proposal contains no such collection of information
requiring OMB approval under the Paperwork Reduction Act.
List of Subjects in 12 CFR Part 1750
Capital classification, Mortgages, Risk-based capital.
Accordingly, for reasons set forth in the preamble, the Office of
Federal Housing Enterprise Oversight proposes to amend 12 CFR part 1750
as follows:
PART 1750--CAPITAL
1. The authority citation for part 1750 as published at 61 FR
29619, June 11, 1996, continues to read as follows:
Authority: 12 U.S.C. 4513, 4514, 4611, 4612, 4614, 4618.
Sec. 1750.5 [Removed]
2. Remove Sec. 1750.5.
3. Amend Sec. 1750.12 of part 1750 as published at 61 FR 29620,
June 11, 1996, by revising paragraph (a) to read as follows:
Sec. 1750.12 Procedures and Timing.
(a) Each Enterprise shall file with the Director a risk-based
capital report each quarter, or at such other times as the Director
requires. The report shall contain information identified by OFHEO in
written instructions to each Enterprise.
* * * * *
4. Revise the Appendix to subpart B of part 1750 as published at 61
FR 29621, June 11, 1996, to read as follows:
Appendix to Subpart B of Part 1750--Risk-Based Capital Test
Methodology and Specifications
1.0 Identification of the Benchmark Loss Experience
1.1 Definitions
1.2 Data
1.3 Procedures
2.0 Identification of a New Benchmark Loss Experience
3.0 Computation of Risk-Based Capital Level
3.1 Enterprise Data
3.1.1 Overview
3.1.2 Whole Loans
3.1.2.1 Characteristics Used to Create Loan Groups
3.1.2.2 Loan Group Characteristics
3.1.2.3 Individual Loan Data
3.1.2.4 Single Family Mortgage Portfolio-Wide Information
3.1.3 Mortgage-Related Securities
3.1.3.1 Single Class MBS Issued by the Enterprises and Ginnie
Mae
3.1.3.2 Derivative Mortgage Securities Issued by the
Enterprises and Ginnie Mae
3.1.3.3 Mortgage Revenue Bonds and Miscellaneous Mortgage-
Related Securities
3.1.4 Non-Mortgage Financial Instruments
3.1.5 Operations, Taxes, and Accounting
3.1.5.1 Data Required to Calculate Taxes, Operating Expenses,
and Dividends
3.1.5.2 Balance Sheet as of the Start of the Stress Test
3.1.6 Other Off-Balance-Sheet Guarantees
3.2 Commitments
3.2.1 Overview
3.2.2 Inputs
3.2.2.1 Loan data
3.2.2.2 Interest Rate Data
3.2.3 Procedures
3.2.4 Output
3.3 Interest Rates
3.3.1 Overview
3.3.2 Inputs
3.3.3 Procedures
3.3.3.1 Identify Starting Values
3.3.3.2 Project the Ten-Year CMT
3.3.3.3 Project the Ten Other CMTs
3.3.3.4 Project Non-Treasury Interest Rates
3.3.3.5 Project Borrowing Rates
3.3.4 Output
3.4 Property Valuation
3.4.1 Overview
3.4.2 Inputs
3.4.3 Procedures
3.4.4 Output
3.5 Mortgage Performance
3.5.1 General
3.5.2 Single Family Default and Prepayment
3.5.2.1 Overview
3.5.2.2 Inputs
3.5.2.3 Procedures
3.5.2.4 Output
3.5.3 Single Family Loss Severity
3.5.3.1 Overview
3.5.3.2 Inputs
3.5.3.3 Procedures
3.5.3.4 Output
3.5.4 Multifamily Default and Prepayment
[[Page 18219]]
3.5.4.1 Overview
3.5.4.2 Inputs
3.5.4.3 Procedures
3.5.4.4 Output
3.5.5 Multifamily Loss Severity
3.5.5.1 Overview
3.5.5.2 Inputs
3.5.5.3 Procedures
3.5.5.4 Output
3.6 Other Credit Factors
3.6.1 Overview
3.6.2 Input
3.6.3 Procedures
3.6.3.1 Identifying Other Credit Factors
3.6.3.2 Classifying Rating Categories in the Stress Test
3.6.3.3 Accounting for Other Credit Factors
3.6.4 Output
3.7 Mortgage Credit Enhancements
3.7.1 Overview
3.7.2 Inputs
3.7.2.1 Enterprise Data on Mortgage Credit Enhancements
3.7.2.2 Public Rating Information
3.7.2.3 Counterparty Coverage Reduction Information
3.7.3 Procedures
3.7.3.1 Classification of Credit Enhancements
3.7.3.2 Calculating Percentage Coverage and Dollar Coverage
Amounts:
3.7.3.3 Calculating Percent of UPB Covered by Each Counterparty
Rating Category
3.7.3.4 Calculating the Percent of UPB Under Dollar-Denominated
Coverage
3.7.3.5 Calculating Coverage Against Credit Losses
3.7.4 Output
3.8 Other Off-Balance Sheet Guarantees
3.8.1 Overview
3.8.2 Input
3.8.3 Procedures
3.8.4 Output
3.9 Cash Flows
3.9.1 Whole Loans
3.9.1.1 Overview
3.9.1.2 Inputs
3.9.1.3 Procedures
3.9.1.4 Output
3.9.2 Mortgage-Related Securities
3.9.2.1 Overview
3.9.2.2 Inputs
3.9.2.3 Procedures
3.9.2.4 Outputs
3.9.3 Debt and Related Cash Flows
3.9.3.1 Overview
3.9.3.2 Inputs
3.9.3.3 Procedures
3.9.3.4 Output
3.9.4 Non-Mortgage Investment and Investment-Linked Derivative
Contract Cash Flows
3.9.4.1 Overview
3.9.4.2 Inputs
3.9.4.3 Procedures
3.9.4.4 Output
3.10 Operations, Taxes, and Accounting
3.10.1 Overview
3.10.2 Inputs
3.10.2.1 Enterprise Data
3.10.2.2 Interest Rates
3.10.2.3 Outputs From Cash Flow Components of the Stress Test
3.10.3 Procedures
3.10.3.1 New Debt and Investments
3.10.3.2 Dividends
3.10.3.3 Allowances for Loan Losses and Other Charge-Offs
3.10.3.4 Operating Expenses
3.10.3.5 Taxes
3.10.3.6 Accounting
3.10.4 Output
3.11 Treatment of New Enterprise Activities
3.12 Calculation of the Risk-Based Capital Requirement
3.12.1 Overview
3.12.2 Inputs
3.12.3 Procedures
3.12.4 Output
1.0 Identification of the Benchmark Loss Experience
OFHEO will use the definitions, data, and methodology described
below to identify the benchmark loss experience.
1.1 Definitions
The terms defined at Sec. 1750.11 shall apply for this Appendix.
In addition, the term Origination year means the year in which a
loan is originated.
1.2 Data
[a] OFHEO identifies the benchmark loss experience using
historical loan-level data required to be submitted by each of the
two Enterprises. OFHEO's analysis is based entirely on the most
current data available on conventional, 30-year, fixed-rate loans
secured by first liens on single-unit, owner-occupied, detached
properties. Detached properties are defined as single family
properties excluding condominiums, planned urban developments, and
cooperatives. The data includes only loans that were purchased by an
Enterprise within 12 months after loan origination and loans for
which the Enterprise has no recourse to the lender.
[b] OFHEO organizes the data from each Enterprise to create two
substantially consistent data sets. OFHEO separately analyzes
default and severity data from each Enterprise. Default rates are
calculated from loan records meeting the criteria specified above.
Severity rates are calculated from the subset of defaulted loans for
which loss data are available.
1.3 Procedures
1.3.1 Cumulative 10-year default rates for each combination of
states and origination years (state/year combination) that OFHEO
examines are calculated for each Enterprise by grouping all of the
Enterprise's loans originated in that combination of states and
years. For origination years with less than 10 years of loss
experience, cumulative-to-date default rates are used. The two
Enterprise default rates are averaged, yielding an ``average default
rate'' for that state/year combination.
1.3.2 An ``average severity rate'' for each state/year
combination is determined in the same manner as the average default
rate. For each Enterprise, the aggregate severity rate is calculated
for all loans in the relevant state/year combination and the two
Enterprise severity rates are averaged.
1.3.3 The ``loss rate'' for any state/year combination examined
is calculated by multiplying the average default rate for that
state/year combination by the average severity rate for that
combination.
1.3.4 The default and severity behavior of loans in the state/
year combination containing at least 2 consecutive origination years
and contiguous areas with a total population equal to or greater
than 5 percent of the population of the United States with the
highest loss rate constitutes the benchmark loss experience.
2.0 Identification of a New Benchmark Loss Experience
OFHEO will periodically monitor available data and reevaluate
the benchmark loss experience using the methodology set forth in
this Appendix. Using this methodology, OFHEO may identify a new
benchmark loss experience that has a higher rate of loss than the
benchmark experience identified at the time of the issuance of this
regulation. In the event such a benchmark is identified, OFHEO may
incorporate the resulting higher loss rates in the stress test.
3.0 Computation of Risk-Based Capital Level
3.1 Enterprise Data
3.1.1 Overview
[a] The stress test requires data on all of an Enterprise's
assets, liabilities, stockholders equity, and off-balance sheet
obligations, as well as the factors that affect them: interest
rates, house prices, rent growth rates, and vacancy rates. This
section characterizes proprietary data of the Enterprises (as
opposed to publicly available data) that are necessary for the
stress test, which are primarily data on Enterprise portfolios of
financial instruments and guarantees as of the start of the stress
test. Data available from public sources that are also necessary for
the stress test--e.g., historical interest rates, house price growth
rates, and public securities data \1\--are described in the sections
of this Appendix that describe the related components of the stress
test (e.g., the Interest Rate component). The stress test uses
proprietary and public data directly, and also uses values derived
from such data. The derivation of these additional values are also
explained in sections of this Appendix. All data as of the start of
the stress test, proprietary data of the Enterprises and public
data, are ``starting position data.''
---------------------------------------------------------------------------
\1\ Data elements listed below for non-mortgage financial
instruments are available from public sources for publicly traded
securities, but are proprietary for privately placed instruments, in
particular, derivative contracts.
---------------------------------------------------------------------------
[b] Starting position data include, for all the loans owned or
guaranteed by an Enterprise, as well as securities and derivative
contracts, the dollar balances of these instruments and obligations,
as well as all characteristics that bear on their behavior under
stress test conditions. Data are required for the following
categories of instruments and obligations:
Mortgages owned by or underlying mortgage-backed
securities issued by the Enterprises (``whole loans'')
Mortgage-related securities
[[Page 18220]]
Non-mortgage-related securities, whether issued by an
Enterprise, e.g., debt, or held as investments
Derivative contracts
Other off-balance sheet guarantees (e.g., guarantees of
private-issue securities)
[c] The stress test also requires starting position data for
``non-cash'' balance sheet items, such as premiums and discounts,
that affect pro forma financial statements through the ten-year
stress period.
3.1.2 Whole Loans
[a] Whole loans are individual single family or multifamily
mortgage loans. The stress test distinguishes between whole loans
that the Enterprises hold in their investment portfolios (retained
loans) and those that underlie mortgage-backed securities (sold
loans). Data are aggregated for loans with similar portfolio
(retained or sold), risk, and product characteristics. The
characteristics of these ``loan groups'' determine mortgage default,
prepayment, and loss severity rates, and cash flows.
[b] The characteristics that are the basis for loan groupings
are called ``classification variables'' and reflect categories,
e.g., fixed interest rate versus floating interest rate, or identify
a value range, e.g., original loan-to-value ratio greater than 80
percent and less than or equal to 90 percent. After the loans are
grouped, weighted average values for characteristics of the loan
group are calculated, e.g., weighted average loan coupon (WAC) and
weighted average remaining maturity (WAM). Loan group
characteristics are used as inputs in section 3.5, Mortgage
Performance, of this Appendix to determine mortgage performance
(default, prepayment, and loss severity) and mortgage cash flows.
3.1.2.1 Characteristics Used to Create Loan Groups
[a] Loan groups are formed based on the values, as of the start
of the stress test, of the relevant loan classification variables
shown in Table 3-1.
[[Page 18221]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.245
[[Page 18222]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.246
[[Page 18223]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.247
[[Page 18224]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.248
[b] All loans with the same values for each of the relevant
characteristics included in Table 3-1 above comprise a single loan
group; for example, one loan group would include all loans with the
following characteristics:
Single family
Sold portfolio
30-year fixed-rate conventional
Originated in 1997
LTV greater than 75 percent and less than or equal to
80 percent
Original coupon greater than or equal to six percent
and less than seven percent
Starting coupon (coupon at the start of the stress
period) greater than or equal to six percent and less than seven
percent
Secured by property located in the East North Central
Census division
Subject to a remittance cycle where scheduled principal
and interest payments are held for an average of seven days
3.1.2.2 Loan Group Characteristics
In addition to the classification variables used for grouping
loans, the stress test requires values for characteristics
calculated for the loans within each group. All values are as of the
start of the stress test. Except as indicated in the ``Description''
column, values are averages for the loans comprising a loan group,
weighted by their unpaid principal balances (UPB).
[[Page 18225]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.249
[[Page 18226]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.250
3.1.2.3 Individual Loan Data
The stress test requires data for individual loans in an
Enterprise's portfolio in order to determine the characteristics of
loans that (for purposes of the stress test) fulfill commitments
that are outstanding at the start of the stress period, and to
compute loss coverage provided by credit enhancements such as
private mortgage insurance. These data requirements are listed
below.
3.1.2.3.1 Commitments Data
[a] To establish the characteristics of loans that fulfill
commitments so that they are consistent with the characteristics of
loans securitized by an Enterprise that were recently originated,
data are required for loans that meet the following criteria:
Single family
Originated within six months of the start date of the
stress test
Securitized
One of the following product types:
1. 30-year fixed-rate
2. 15-year fixed-rate
3. One-year CMT ARM
4. Seven-year balloon
[b] For these loans, the following data are required:
Loan balance as of the beginning of the stress period
Original LTV
Census division
Guarantee fee
Servicing fee
Margin (for ARM loans)
Credit enhancement data described in section 3.1.2.3.2,
Credit Enhancement Data, below
[c] The dollar amount of commitments outstanding at the start of
the stress test is also required.
3.1.2.3.2 Credit Enhancement Data
[a] To facilitate calculation of the reductions in mortgage
credit losses due to credit enhancements, the following data are
required for all credit-enhanced loans, if any, in a loan group:
1. Type of mortgage credit enhancement:
a. Private mortgage insurance
b. Recourse
Limited
Unlimited
c. Indemnification
Limited
Unlimited
d. Pool insurance
e. Spread account
f. Collateral posted under collateral pledge agreement
g. Cash account
2. Private mortgage insurance coverage percent
3. Loan balance as of the beginning of the stress period
4. Public rating of mortgage insurer
5. Public rating of pool insurer
6. Public rating of seller or servicer
[b] The following additional information is needed for each loan
delivery contract involving a spread account, collateral account,
cash, limited recourse or indemnification, or pool insurance account
(e.g., a particular contract for the delivery of $100 million of
loans may specify the establishment of a spread account as credit
enhancement):
Coverage remaining, as of the beginning of the stress
period
Account balance(s) at the start of the stress period
Coverage expiration date
3.1.2.4 Single Family Mortgage Portfolio-Wide Information
To reflect the differential performance of single family
mortgages on investor-owned and owner-occupied properties, the
stress test also requires data on the percentage of first lien
mortgages in the combined retained and sold portfolios financing
investor-owned properties.
3.1.3 Mortgage-Related Securities
[a] The Enterprises hold mortgage-related securities as assets.
These securities include single class and derivative mortgage-backed
securities (multi-class and strip securities) issued by Fannie Mae,
Freddie Mac, and Ginnie Mae; mortgage revenue bonds issued by State
and local governments and their instrumentalities; and single class
and derivative mortgage-backed securities issued by private
entities. Most mortgage-related securities are collateralized by
single family mortgages, others by multifamily mortgages, and, for
the purposes of the stress test, still others by housing-related
assets such as manufactured housing loans.
[b] The stress test models the cash flows of these securities
individually. Enterprise data required for this purpose are
described below.
3.1.3.1 Single Class MBS Issued by the Enterprises and Ginnie Mae
[a] Table 3-3 provides Enterprise data regarding each MBS held
in their portfolios. This information is necessary for simulating
cash flows in the stress test.
[[Page 18227]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.251
[b] The Enterprises and Ginnie Mae make available to the public
monthly pool data that provide investors with information on
principal payments, as well as extensive data characterizing
individual MBS and their underlying mortgage pools. These data,
which are necessary to simulate MBS cash flows, are listed in
section 3.9.2, Mortgage-Related Securities, of this Appendix.
3.1.3.2 Derivative Mortgage Securities Issued by the Enterprises and
Ginnie Mae
[a] Table 3-4 provides Enterprise data regarding REMICs and
Strips issued by the Enterprises or Ginnie Mae. This information is
necessary for determining associated cash flows.
[GRAPHIC] [TIFF OMITTED] TP13AP99.252
[b] The data in Table 3-4 identify individual securities that
are held by the Enterprises in their portfolios, as well as the
REMIC or Strip transaction associated with individual securities.
Public securities disclosure information is the source of data on
the collateral underlying the securities (e.g., pool numbers of
securities comprising collateral for a series of securities) and the
rules governing security cash flows. (See section 3.9.2, Mortgage-
Related Securities, of this Appendix.)
3.1.3.3 Mortgage Revenue Bonds and Miscellaneous Mortgage-Related
Securities
[a] Table 3-5 provides Enterprise data regarding mortgage
revenue bonds and private-issue, mortgage-related securities (MRS).
This information is necessary for determining associated cash flows.
[GRAPHIC] [TIFF OMITTED] TP13AP99.253
[[Page 18228]]
[b] The data in Table 3-5 are supplemented with public
securities disclosure data, as described in section 3.9.2, Mortgage-
Related Securities, of this Appendix.
3.1.4 Non-Mortgage Financial Instruments
[a] Non-mortgage financial instruments include debt securities
issued to fund assets, debt securities and preferred stock held as
assets, derivatives contracts (interest rate swaps, caps, and
floors), and preferred stock issued by an Enterprise. Cash flows for
non-mortgage financial instruments are simulated based on their
characteristics. Although information for publicly traded
securities, including most of the Enterprises' debt securities and
non-mortgage investments, is available from public securities
disclosure documents, information on other derivative contracts and
non-publicly traded instruments must be obtained from the
Enterprises. Data categories listed here apply to both publicly
traded and privately placed instruments. All data are instrument
specific; the pay- and receive-sides of swap contracts are treated
as separate instruments. Table 3-6 provides basic information about
non-mortgage financial instruments input variables, as follows:
[GRAPHIC] [TIFF OMITTED] TP13AP99.254
[[Page 18229]]
[b] Occasionally, instruments have complex or non-standard
features, and cash flows cannot be computed using the basic data
listed above. In these cases the accurate modeling of cash flows
requires additional information, such as amortization schedules,
interest rate coupon reset formulas, and the terms of European call
options, which is obtained from the Enterprises (and is included in
public securities disclosure materials for publicly offered
securities).
3.1.5 Operations, Taxes, and Accounting
The stress test determines how much total capital an Enterprise
must hold at the start of the stress test so that total capital
never falls below zero during the stress period. To accomplish this
objective, projected cash flows for Enterprise financial instruments
must be supplemented by projected operating expenses, taxes, and
capital distributions. All of these must be recorded in pro forma
financial statements in order to determine an Enterprise's total
capital for each month of the stress period. Thus, complete
information on the Enterprise balance sheet as of the start of the
stress period is required. The necessary information is listed in
section 3.1.5.1, Data Required to Calculate Taxes, Operating
Expenses, and Dividends, below.
3.1.5.1 Data Required to Calculate Taxes, Operating Expenses, and
Dividends
The following Enterprise data are necessary to calculate taxes,
operating expenses, and dividends:
Operating expenses (e.g., administrative expenses,
salaries and benefits, professional services, property costs,
equipment costs) for the quarter prior to the beginning of the
stress test
Earnings before income taxes and provision for income
taxes for the three years prior to the beginning of the stress
period
Year-to-date income before taxes and provision for
income taxes
Dividend payout ratio for the four quarters prior to
the beginning of the stress period
Minimum capital requirement as of the beginning of the
stress period
3.1.5.2 Balance Sheet as of the Start of the Stress Test
The data are necessary to create Enterprise balance sheets as of
the start of the stress period are described below.
1. Balances for all instruments for which the stress test
calculates cash flows. These are included with data the Enterprises
provide for cash flow calculations. Balances are required for:
Whole loans
Mortgage-related securities
Non-mortgage investments and investment-linked
derivative contracts
Debt and related cash flows
2. Additional starting position balances:
Amounts required to reconcile starting position
balances from cash flow components of the stress test with an
Enterprise's balance sheet (for example, differences between actual
and estimated loan prepayments during the last few days in the
month)
Cash
Low income housing tax credit investments
Unamortized balances of premiums, discounts, and fees
from the acquisition of retained loans and mortgage-related
securities at other than par value
Allowances for loan losses
Accrued interest receivable on retained loans,
mortgage-backed securities, mortgage-linked derivatives, and non-
mortgage investments
Amounts receivable from Index Sinking Fund Debentures,
currency swaps, fees, income taxes, and other accounts receivable
Real estate owned (REO)
Fixed assets
Clearing accounts
Unamortized premiums, discounts, and fees related to
debt securities
Unamortized balances related to the sold portfolio
Deferred balances related to liability-linked
derivatives
Accrued interest payable
Principal and interest payable to mortgage security
investors
Other liabilities, including payables from currency
swaps, escrow deposits income taxes
Dividends payable
Components of stockholder's equity (i.e., common stock,
preferred stock, paid-in capital, retained earnings, treasury stock,
and unrealized gains and losses on available-for-sale securities)
3.1.6 Other Off-Balance-Sheet Guarantees
In addition to the MBS they issue, the Enterprises guarantee
other securities. The stress test does not simulate the cash flows
associated with these guarantees, but it does calculate an
incremental capital requirement for them. This calculation requires
Enterprise information on the sum of the outstanding balances of all
tax-exempt multifamily housing bonds, single-family whole-loan
REMICs, multifamily whole-loan REMICs, and similar instruments or
obligations as of the beginning of the stress period (excluding all
guarantees of securities where 100 percent of collateral is insured
by FHA or guaranteed by VA).\2\
---------------------------------------------------------------------------
\2\ These include: (1) Any guarantee, pledge, purchase
arrangement, or other obligation or commitment provided or entered
into by an Enterprise with respect to multifamily mortgages to
provide credit enhancement, liquidity, interest rate support, and
other guarantees and enhancements for revenue bonds issued by a
state or local government unit (including a housing finance agency)
or other bond issuer; and (2) all off-balance-sheet obligations of
an Enterprise that are not mortgage-backed securities or
substantially equivalent instruments and that are not resecuritized
mortgage-backed securities, such as real estate mortgage investment
conduits or similar resecuritized instruments. See 12 CFR 1750.2.
---------------------------------------------------------------------------
3.2 Commitments
3.2.1 Overview
The Enterprises make contractual commitments to their customers
to purchase or securitize mortgages. The stress test provides for
deliveries of mortgages under the commitments that exist at the
start of the stress period. It also determines all of the relevant
characteristics of these mortgages by reference to the
characteristics of the mortgages securitized by the Enterprise that
were originated in the six months preceding the start of the stress
period. Based on this information, the Commitments component of the
stress test creates loan groups with coupon rates that vary based
upon the interest rate scenario. These loan groups are added to the
Enterprise's sold portfolio and the stress test projects their
performance during the stress period. In the down-rate scenario, the
stress test provides that 100 percent of the mortgages specified in
the commitments are delivered. In the up-rate scenario, 75 percent
are delivered. Loans are delivered over the first three months of
the stress period in the down-rate scenario and the first six months
in the up-rate scenario.
3.2.2 Inputs
The stress test uses two sources of data to determine the
characteristics of the mortgages delivered under commitments. One is
information from the Enterprises on commitments outstanding at the
start of the stress period and deliveries of loans originated in the
six months preceding the start of the stress period (See section
3.1.2, Whole Loans, of this Appendix). The other is interest rate
series generated by the Interest Rates component of the stress test
(See section 3.3, Interest Rates, of this Appendix).
3.2.2.1 Loan Data
[a] To determine the total dollar amount of mortgages that will
be delivered under commitments during the course of the stress
period, the Enterprises are required to provide the total dollar
amount of all commitments outstanding to purchase or securitize
mortgages at the start of the stress period. In addition, to
determine the composition of mortgages delivered to fulfill
commitments, the stress test identifies loans that meet all of the
following criteria:
Business type-single family
Origination date-within six months of the start date of
the stress test
Portfolio type-securitized
Product type-one of the following:
1. 30-year fixed-rate
2. 15-year fixed-rate
3. One-year CMT ARM
4. Seven-year balloon
[b] For the selected loans, the following loan-level information
are required:
Starting UPB
Original LTV
Census division
Guarantee fee
Margin (for ARM loans)
Servicing fee
3.2.2.2 Interest Rate Data
The stress test uses the following interest rate series,
generated by the Interest Rates component, (See section 3.3,
Interest Rates, of this Appendix) for the first 12 months of the
stress period:
One-year CMT rate
Conventional 30-year fixed-rate mortgage rate
Conventional 15-year fixed-rate mortgage rate
Seven-year balloon mortgage rate \3\
---------------------------------------------------------------------------
\3\ The stress text assumes that mortgage interest rates on
seven-year balloon mortgages are 50 basis points less than 30-year
conventional mortgage rates in the down-rate environment, and equal
to the 30-year rate in the up-rate environment.
---------------------------------------------------------------------------
[[Page 18230]]
3.2.3 Procedures
[a] Based on the characteristics of the mortgages securitized by
the Enterprise that were originated in the six months preceding the
start of the stress period and the interest rate projections in the
stress period, the stress test determines all of the relevant
characteristics of the loans delivered under the commitments that
exist at the start of the stress test. Using this information and
the classification variables-business type, portfolio type, product
type, original loan-to-value ratio, and Census division, the stress
test creates loan groups for commitments in the same manner as loan
groups are created for other loans (specified in section 3.1.2,
Whole Loans, of this Appendix). One exception is that the stress
test uses an additional classification variable--delivery month--to
form subgroups within each commitment loan group. This variable is
used to create origination dates, which are the same as delivery
dates for these loan groups. The procedures to create commitment
loan groups are as follows.
1. Establish the values for classification variables--business
type, portfolio type, product type, original loan-to-value ratio,
and Census division as defined in section 3.1, Enterprise Data, of
this Appendix.
2. Aggregate the loan-level information for the mortgages
identified above into loan groups by the classification variables.
3. Concurrently with step 2, compute total starting UPB, the UPB
weighted average Original LTV, Servicing fee, Guarantee fee, and
Margin (for ARM loans) for each loan group.
4. Using loan group information from step 3, calculate the
percent of total balance of all commitment loan groups for each loan
group as follows:
% of total balance = total starting UPB for the loan group (from
step 3 above) total starting UPB for all commitment loan
groups added together
5. For each loan group, set the loan term and amortization
period as shown in Table 3-7.
[GRAPHIC] [TIFF OMITTED] TP13AP99.255
6. For each loan group, set remittance cycle to the shortest
available option for the Enterprise.
[b] Procedures for adding subgroup characteristics to each loan
group are described below.
1. Establish values for the subgroup classification variable--
delivery month using percentages from Table 3-8, and divide each
loan group into subgroups, one for each delivery month. Three
subgroups are created in the down-rate scenario, and six subgroups
are created in the up-rate scenario.
2. The total starting UPB for the subgroup is calculated as
follows: subgroup balance = total dollar amount of commitments
outstanding x % of total balance of the subgroup (from step 4
above) x Percent delivered in that delivery month (from Table 3-
8).
[GRAPHIC] [TIFF OMITTED] TP13AP99.256
3. Set the original coupon rate and starting coupon rate (as of
delivery date) for each subgroup as set forth in Table 3-9.
[[Page 18231]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.257
4. Based on the original coupon rate and starting coupon rate
set for the subgroup in step 3, assign the subgroup with original
coupon rate class value and starting coupon rate class value as
defined in section 3.1.2, Whole Loans, of this Appendix.
5. Set the origination year and month of the subgroup by adding
the delivery month to the starting date of the stress period.
6. Set the age of the subgroup in the stress period to the
number of months elapsed in the stress period minus the delivery
month. Set the remaining term of the subgroup to the amortization
term minus the age of the subgroup.
7. Set the net yield of the subgroup to the starting coupon rate
minus the servicing fee.
8. Set the passthrough rate of the subgroup to the net yield
minus the guarantee fee.
3.2.4 Output
[a] The output of the Commitment component of the stress test is
data for a set of loan subgroups that are virtually identical to
loan groups created for loans on the books of business of the
Enterprises at the start of the stress test, except that an
additional classification variable, delivery month, is used to
supplement origination year for each subgroup of commitment loans.
This additional information tells when the mortgages in that
particular subgroup are delivered to the Enterprise.
[b] The data for loan subgroups created by the Commitments
component of the stress test allows the stress test to project the
defaults, losses, prepayments, scheduled amortization, interest
payments, guarantee fee income, and float income for loans purchased
under commitments for the ten-year stress period.
3.3 Interest Rates
3.3.1 Overview
The 1992 Act specifies changes in the ten-year constant maturity
Treasury yield (CMT) for the two interest rate scenarios of the
stress test. It further states that yields of Treasury instruments
with other maturities will change relative to the ten-year CMT in
patterns that are reasonably related to historical experience. The
Interest Rates component of the stress test projects these Treasury
yields as well as other interest rate indexes that are needed to
calculate cash flows, to simulate mortgage performance for mortgages
and other financial instruments, and to calculate the risk-based
capital requirement. The Interest Rates component produces values
for the interest rates and indexes for the starting date of the
stress test and for each of the 120 months in the stress period. The
process for determining interest rates can be divided into five
steps. First, identify values for the necessary interest rates and
indexes on the starting date. Second, project the ten-year CMT for
each month of the stress period as specified in the 1992 Act. Third,
project the one-, two-, three-, and six-month Treasury yields and
the one-, two-, three-, five-, 20-and 30-year CMTs.\4\ Fourth,
project non-Treasury indexes and interest rates. Fifth, project
borrowing rates for the Enterprises.
---------------------------------------------------------------------------
\4\ For ease of discussion, all of the Treasury yields are
referred to as CMTs.
---------------------------------------------------------------------------
3.3.2 Inputs
Projecting interest rates and indexes in the stress test
requires initial values as of the start date of the stress test.
Initial values for the stress test are the averages of the values
for the month preceding the start of the stress period. Additional
months of historical data are input to the stress test in order to
project interest rates other than the ten-year CMTs during the
stress period. The historical data input for non-Treasury interest
rate indexes are listed in Table 3-12. Table 3-10 below contains a
list and a description of the interest rates and indexes input to
the stress test.
[[Page 18232]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.258
[[Page 18233]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.259
3.3.3 Procedures
3.3.3.1 Identify Starting Values
The starting values for all of the interest rates and indexes
listed in Table 3-10 are their daily averages during the month
preceding the start of the stress test.
3.3.3.2 Project the Ten-Year CMT
The 1992 Act specifies that the stress test be based on
increases or decreases in the ten-year CMT, whichever would require
more capital. The ten-year CMT increases or decreases during the
first year of the stress period and remains at that level for the
remainder of the stress period. The 1992 Act further specifies how
the increases and decreases in the ten-year CMT are determined.
3.3.3.2.1 Down-Rate Scenario
[a] To determine the ten-year CMT in the down-rate scenario, the
stress test first computes the average of the ten-year CMT for the
nine months prior to the start of the stress test, and subtracts 600
basis points; and second, computes the average yield of the ten-year
CMT for the 36 months prior to the start of the stress test, and
multiplies by 60 percent.
[b] The ten-year CMT in the down-rate scenario is decreased to
the lesser of these two yields unless that yield is less than 50
percent of the average for the nine months preceding the start date.
In that case, the ten-year CMT decreases 50 percent of the nine-
month average described above.
[c] Once the ten-year CMT for the down-rate scenario is
determined, the stress test decreases the ten-year CMT from the
value as of the start of the stress period to this level in equal
increments over the first twelve months of the stress period. The
ten-year CMT remains at this level for the remaining nine years of
the stress period.
3.3.3.2.2 Up-Rate Scenario
[a] To determine the ten-year CMT in the up-rate scenario, the
stress test first computes the average for the ten-year CMT the nine
months prior to the start of the stress test, and adds 600 basis
points; and second, computes the average for the ten-year CMT for
the 36 months prior to the start of the stress test, and multiplies
by 160 percent.
[b] The ten-year CMT in the up-rate case is equal to the greater
of these two rates unless that yield is greater than 175 percent of
the average for the nine months preceding the stress period. In that
case, the ten-year CMT increases to 175 percent of the nine-month
average.
[c] Once the ten-year CMT for the up-rate scenario is
determined, the stress test increases the ten-year CMT from the
value as of the start of the stress period to this level in equal
increments over the first twelve months of the stress period. The
ten-year CMT remains at this level for the remaining nine years of
the stress period.
3.3.3.3 Project the Ten Other CMTs
In the third step, yields for the one-, two-, three-and six-
month and the one-, two, three-, five-, 20-and 30-year CMTs are
projected.
3.3.3.3.1 Down-Rate Scenario
[a] In the down-rate scenario, the ten other CMTs are calculated
by first computing the long-term averages for the ten-year CMT and
each of the ten CMTs, and then computing the ratios of the ten-year
CMT long-term average to the ten other CMT long-term averages. The
long-term averages are calculated over the period from May, 1986,
through April, 1995. These are presented in Table 3-11 below. The
stress test multiplies the ten-year CMT for the last nine years of
the stress test by the appropriate ratio to create the six other
CMTs for the last nine years of the stress test.
[[Page 18234]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.260
[b] In the first twelve months of the stress period, the ten
other CMTs are computed in a manner similar to the calculation of
the ten-year CMT for that period. From its value at the start of the
stress test, each of the ten other CMTs is decreased in equal steps
in each of the first twelve months of the stress period until it
reaches the appropriate level for the nine remaining years of the
stress test.
3.3.3.3.2 Up-Rate Scenario
In the up-rate scenario, the six other CMTs are equal to the
ten-year CMT in the last nine years of the stress test. Each of the
six other CMTs is increased in equal increments over the first
twelve months of the stress test until it equals the ten-year CMT.
3.3.3.4 Project Non-Treasury Interest Rates
[a] Table 3-12 presents the equations for projecting the non-
Treasury interest rates for each month of the stress test. These
equations were developed using the percentage spread between the
non-Treasury interest rate and the CMT with the same or similar
maturity over a historical period \5\ and an ARIMA procedure
(Autoregressive Integrated Moving Average).\6\ The stress test
applies these equations to forecast the spreads between each non-
Treasury interest rate and the CMT from which it is estimated for
the 120 months of the stress period. Finally, the stress test
converts the projected values for the proportional spreads into rate
and index levels. As used here, the percentage spread for the three-
month LIBOR rate, for example, is:
---------------------------------------------------------------------------
\5\ Various historical data series have missing values.
\6\ SAS ETS Users Guide, SAS Institute, 1993.
[GRAPHIC] [TIFF OMITTED] TP13AP99.062
[b] In Table 3-12, equations are grouped according to the
Treasury maturity against which the spread was calculated. For
example, the first group's spread was computed against the one-month
Treasury yield. Where the dependent variable was estimated as a
first difference, this is indicated in the Description column. ``T''
represents the spread variable.
[[Page 18235]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.261
[[Page 18236]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.262
3.3.3.5 Project Borrowing Rates
The stress test adds a 50 basis point credit spread to the
federal agency cost of funds index to project Enterprise borrowing
costs for the last nine years of the stress period.
3.3.4 Output
The output from the interest rate calculations are 120 monthly
interest rate and index values for the projected eleven points on
the Treasury yield curve (one-month, two-month, three-month, six-
month, one-year, two-year, three-year, five-year, ten-year, 20-year
and 30-year) and the 20 non-Treasury yields.
3.4 Property Valuation
3.4.1 Overview
[a] The Property Valuation component provides the monthly single
family house price growth rates, rent growth rates, and rental unit
vacancy rates that contribute to the determination of property
values in the calculation of mortgage performance. The rates are
those associated with the benchmark loss experience, the ten-year
experience of loans originated in Arkansas, Louisiana, Mississippi,
and Oklahoma during 1983 and 1984. The benchmark loss experience
spans twelve years from the beginning of 1983, when the first
benchmark loans were originated, through the end of 1994, ten years
after the last benchmark loans were originated. The rates used in
the stress test are those for the middle ten years of this period,
1984 through 1993.
[b] Single family house price growth rates are taken from the
HPI series for the West South Central Census Division, which
includes all of the benchmark states except Mississippi. House price
growth rates are used to project single family mortgage performance.
Rent growth rates and vacancy rates are taken from information for
the major metropolitan areas in the four benchmark States, published
by the Institute for Real Estate Management, and State level vacancy
rates published by the Bureau of the Census. These rates are used to
project multifamily mortgage performance.
[c] As required by the 1992 Act, in the up-rate scenario, house
price rates and rent growth rates may require adjustment for
inflation. If the ten-year CMT rises more than 50 percent from the
average yield during the nine months preceding the stress period,
rates are adjusted upward to take into account the effect of
inflation.
[d] This section includes a description of the required inputs
and procedures for inflation adjustments, and concludes with
outputs. These outputs include tables of benchmark house price and
rent growth rates unadjusted for inflation and rental vacancy rates.
These rates will not change unless the benchmark loss experience
changes.
3.4.2 Inputs
The inputs required for adjusting house price and rent growth
rates are:
The average yield of the ten-year CMT for the nine
months preceding the stress period, as computed in section 3.3,
Interest Rates, of this Appendix)
The highest 10-year CMT during the stress period, as
computed in section 3.3, Interest Rates, of this Appendix
[[Page 18237]]
Unadjusted house price and rent growth rates during the
stress period, as shown in Tables 3-13 and 3-14 below
3.4.3 Procedures
Inflation adjustments are applied over the final five years of
the up-rate scenario stress test. The procedures are described
below.
1. Determine whether an adjustment is necessary. Multiply the
average10-year CMT for the nine months preceding the stress period
by 1.50, and subtract the product from the highest value of the10-
year CMT during the stress period. The difference is YD. If YD > 0,
follow steps 2-4 to apply an inflation adjustment. Otherwise, use
the rates provided in the Tables 3-13 and 3-14.
2. Compute the adjustment. Use the following formula to compute
the cumulative adjustment as if YD were to apply over 9 years and 2
months: \7\
---------------------------------------------------------------------------
\7\ If the ten-year CMT increases 75 percent over the base
month, a 50 percent increase will be achieved by month eight. The
full increase will be achieved by month twelve. On average, the
difference YD will apply for 9 years and 2 months.
[GRAPHIC] [TIFF OMITTED] TP13AP99.063
[GRAPHIC] [TIFF OMITTED] TP13AP99.287
3. Calculate the monthly inflation adjustment factors to apply
to house price and rent rate growth rates. The cumulative adjustment
is applied over the last five years of the stress period, and
monthly adjustment factors are computed as follows:
a. For house-price growth rates, the monthly adjustment factor
is: \8\
\8\ This factor is in continuous rate form (note use of natural
logarithm) to be compatible with the house price growth rate series
in Table 3-13.
[GRAPHIC] [TIFF OMITTED] TP13AP99.064
[GRAPHIC] [TIFF OMITTED] TP13AP99.288
b. For rent growth rates, the monthly adjustment factor is: \9\
\9\ This factor is in discrete rate form to be compatible with
the rent growth rate series in Table 3-14.
[GRAPHIC] [TIFF OMITTED] TP13AP99.065
[GRAPHIC] [TIFF OMITTED] TP13AP99.289
4. Compute final monthly growth rates. Add the monthly inflation
adjustment factors IHt and IRt to the house
and rent growth rates for months 61 through 120. The resulting
series will be inflation-adjusted growth rates.
3.4.4 Output
[a] Monthly house price growth rates, rent growth rates, and
rental vacancy rates are used by the Mortgage Performance components
of the stress test (see section 3.5, Mortgage Performance, of this
Appendix). If there are no inflation adjustments, the house price
and rent growth rates in Tables 3-13 and 3-14 are used. If the
inflation adjustment is necessary, then the adjusted growth rates
are used.
[b] House price growth rates are inputs to the Single Family
Default and Prepayment and the Single Family Loss Severity
components of the stress test (See sections 3.5.2 and 3.5.3 of this
Appendix). The rent growth rates and vacancy rates are inputs to the
Multifamily Default and Prepayment and Multifamily Loss Severity
components (See sections 3.5.4 and 3.5.5 of this Appendix).
[[Page 18238]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.263
[[Page 18239]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.264
[GRAPHIC] [TIFF OMITTED] TP13AP99.265
[[Page 18240]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.266
3.5 Mortgage Performance
3.5.1 General
[a] The four components of the stress test that simulate various
elements of mortgage performance are single family default and
prepayment, single family loss severity, multifamily default and
prepayment, and multifamily loss severity.
[b] Figure 3-1 is a schematic overview of the basic structure of
each mortgage performance component. Each mortgage performance
component uses as inputs loan group starting position data, interest
rate series from the Interest Rates component (see section 3.3,
Interest Rates, of this Appendix), historical house-price indexes
(HPI) and rental-price indexes (RPI) from government sources, and
HPI and RPI growth and rental vacancy rate series for the stress
period from section 3.4, Property Valuation, of this Appendix. These
inputs are used to calculate the values of explanatory variables
that are then used to compute monthly default, prepayment, and loss
severity rates. These monthly default, prepayment, and loss severity
rates are used to compute cash flows (refer to section 3.9, Cash
Flows, of this Appendix).
[[Page 18241]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.372
3.5.2 Single Family Default and Prepayment
3.5.2.1 Overview
The stress test calculates conditional default and prepayment
rates for single family mortgages for each month of the ten-year
stress period. A conditional rate of default or prepayment refers to
the percentage of the outstanding balance in a loan group that
defaults or prepays during a given period of time. Computing default
and prepayment rates requires information on the risk
characteristics of a loans, historical and projected rates of
interest, and the historical and projected rates of property value
appreciation (or depreciation). Some of this information is used
directly, while other information is combined together to create new
variables for use in the default and prepayment rate calculations.
In all, nine explanatory variables are used to determine default and
prepayment rates for single family loans: mortgage age, mortgage age
squared, original loan-to-value ratio, probability of negative
equity, prepayment burnout, the percentage of investment property
loans, relative interest rate spread, yield curve slope, and
mortgage product-type. A statistical analysis of the relationship
between the explanatory variables and historical default and
prepayment rates was used to estimate the weights (also known as
regression coefficients) associated with each variable. The selected
weights are combined as described below to compute quarterly default
and prepayment rates throughout the stress test period. The
quarterly rates are then converted to monthly conditional default
and prepayment rates and used by the cash flow component (See
section 3.9, Cash Flows, of this Appendix) of the stress test to
calculate monthly principal reductions resulting from defaults and
prepayments, and to calculate default losses for each month in the
ten-year stress period.
3.5.2.2 Inputs
[a] There are three categories of data inputs for single family
default and prepayment rate calculations: characteristics of loan
groups, interest rates, and house price index values and
volatilities.
[b] The loan group characteristics used here are listed below
with their
[[Page 18242]]
corresponding variable names, where relevant, as they appear in
subsequent formulas:
Product type
Origination year (Y0)
Origination month (required for loans delivered under
commitments only)
Census division (d)
Origination LTV (LTV0)
Origination UPB (UPB0)
Original coupon interest rate (rc,0)
Mortgage origination term, in months (To)
Mortgage amortization term, in months (Ta)
Remaining term, in months (Tr)
Percentage of investor loans (P) (this refers to the
percent of investor property loans in an Enterprise's entire loan
portfolio)
[c] The interest rate variables are listed below, along with
their reference names as they appear in subsequent formulas:
Conventional 30-year fixed-rate mortgage coupon rates
(rf,q)
One-year (Constant Maturity) Treasury yields
(y12q)
Ten-year (Constant Maturity) Treasury yields
(y120q)
[d] All interest rate series are provided by the Interest Rate
component in monthly form. They are converted to quarterly series by
taking simple averages of monthly values within each calendar
quarter. Each interest-rate series represents 30 years of historical
values, plus 10 years of stress test values. As described below in
section 3.5.2.3, Procedures, of this Appendix, loans with
origination years prior to 1979 are treated as having an origination
year of 1979. Therefore, no interest rate variable values before
that year are used. The conventional 30-year fixed-rate mortgage
rate series does not begin until the second half of 1979, so values
for the first two quarters of 1979 are equal to the third-quarter
value.
[e] House price growth rates are used to adjust the value of
collateral properties before and during the stress period. Before
the stress test is run, mortgages are seasoned using historical
Census Division HPI series from the most recent OFHEO HPI report.
House price growth rates for the stress period are determined as
discussed in section 3.4, Property Valuation, of this Appendix. The
two house price growth rate volatility parameters published in the
OFHEO HPI Report, for each Census division, are also used, as
described below. The volatility parameters measure the distribution
of individual house price growth paths around the measured HPI
value, as a function of the age of a mortgage.
3.5.2.3 Procedures
3.5.2.3.1 Overview
Five general steps for generating default and prepayment rates
for single family loans are repeated for each loan group throughout
the stress period.
1. Obtain the loan group characteristics, the interest rates,
and the HPI index and volatility values.
2. Using the loan characteristics and other input data, compute
the values for the nine explanatory variables, by loan group, for
each quarter of the stress period.
3. Match the time series of explanatory variables for each loan
group to associated regression weights (coefficients) for use in
calculating default and prepayment rate series. Some of the
variables are multiplied by the weights and then used in the default
and prepayment rate calculations. These are called ``continuous''
variables, and they include age (and age squared), investor-property
percent. Other variables are categorical and do not get multiplied
by the weights. Rather, for these explanatory variables, one of
several available weights is assigned based on the value-range or
category of the explanatory variable value in each quarter. For
categorical variables, the underlying values can change from quarter
to quarter, and the weights used will also change, as the variable
value moves from one category to another.
4. Sum the results of Step 3--a combined set of weighted
continuous variables and categorical variable weights for each
quarter--to produce factors that go into default and prepayment rate
calculations. The rate calculations use logistic probability
formulas. Table 3-17 provides all weights needed to compute the
default and prepayment rates for each product type. There is one set
of beta () and gamma () weights for 30-year fixed-
rate mortgages, one set for adjustable rate mortgages, and one set
for all other product types.
5. Convert the quarter default and prepayment rates into monthly
equivalent rates so that the stress test has monthly series for cash
flow projections.
3.5.2.3.2 Explanatory Variables Calculations
The following sections describe how each explanatory variable is
calculated and how the weights are combined to compute default and
prepayment rates for a group of single family loans of similar risk
characteristics.
3.5.2.3.2.1 Mortgage Age (Aq)
[a] The mortgage age in each quarter of the stress period is
computed as:
[GRAPHIC] [TIFF OMITTED] TP13AP99.066
[GRAPHIC] [TIFF OMITTED] TP13AP99.290
[b] Loans with origination years prior to 1979 are treated as if
they were originated in 1979. The age value and the squared value of
age are used directly in the default and prepayment formula, along
with their weights (coefficients).
3.5.2.3.2.2 Origination LTV (LTV0)
The value of the original LTV for each loan group does not
change throughout the stress test. Once it is matched to an
LTV0 category in Table 3-17, the associated default and
[[Page 18243]]
prepayment weights are used throughout the stress test.\10\
---------------------------------------------------------------------------
\10\ Note that Table 3-1 of this Appendix shows eight categories
for original LTV ratio classes. The default and prepayment component
of the stress test combines the last three categories into one
category.
---------------------------------------------------------------------------
3.5.2.3.2.3 Probability of Negative Equity (PNEQq)
[a] The probability of negative equity variable requires
creating a time series of property values and amortizing loans to
create updated LTV ratios throughout the stress period. The updated
LTV ratios are used along with the standard deviations of house
price growth paths to compute probabilities of negative equity. The
probability of negative equity measures the percent of loans
underlying a loan group that are likely to have negative equity
positions, in each quarter of the stress period. The step-by-step
process for computing the variable PNEQq follows. See
Figure 3-2 for an overview of the derivation process.
1. Create a time series of property values that extends from
loan origination through the stress period as described below.
a. Extend the historical HPI series for each of the nine Census
divisions through the stress period by adding the growth rate
factors (gi) that are described in section 3.4, Property
Valuation, of this Appendix:
[GRAPHIC] [TIFF OMITTED] TP13AP99.067
[GRAPHIC] [TIFF OMITTED] TP13AP99.291
b. Create an index for average house value in each quarter of
the stress period (Vq) using HPI values from the loan
origination quarter and from each quarter of the stress period, by
Census division:
[GRAPHIC] [TIFF OMITTED] TP13AP99.068
[GRAPHIC] [TIFF OMITTED] TP13AP99.292
The published HPI series begins in the first quarter of 1980.
Values for the four quarters of 1979 are produced by OFHEO, but are
not published. Table 3-16 provides these values, which are assigned
to HPId,O for loans originating in 1979. Loans with
origination years prior to 1979 are treated as if they were
originated in 1979.
[[Page 18244]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.267
2. Amortize the average loan balance from loan origination
through the stress period. This procedure does not use the current
mortgage coupon rate at the start of the stress period, but rather
creates a history of interest rate paths for the loan group, from
origination, as if all adjustable rate mortgages are Constant
Maturity Treasury ARMs, with one-year adjustment periods.
a. Create the coupon interest rate series, rc,q. For
fixed-rate mortgages, set rc,q = rc,0,
(original coupon) for every quarter. However, for adjustable-rate
mortgages, adjustments must be made over time, taking into account
period and lifetime interest rate caps as follows:
First, set rc,q = rc,0 for q = {1,...,4}.
[GRAPHIC] [TIFF OMITTED] TP13AP99.293
Then, for every fourth quarter of loan life, evaluate:
[GRAPHIC] [TIFF OMITTED] TP13AP99.069
[GRAPHIC] [TIFF OMITTED] TP13AP99.294
When rc,q<(y12q +0.0275), then set:
rc,q+1...q+4 = min{(y12q + 0.0275),
(rc,q + 0.02), (rc,0 + 0.05)}
When rc,q>(y12q +0.0275), then set:
rc,q+1...q+4 = max{(y12q + 0.0275),
(rc,q-0.02), (rc,0-0.05)}
When rc,q = (y12q +0.0275), then set:
rc,q+1...q+4 = rc,q
[GRAPHIC] [TIFF OMITTED] TP13AP99.295
b. Compute the monthly mortgage payment factor (PMTq)
for each quarter of the stress period, q = {1,...,40} using the
formula:
[[Page 18245]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.070
[GRAPHIC] [TIFF OMITTED] TP13AP99.296
In this formula, LTV0 represents the original loan
balance. Using LTV0 allows the UPB time series to be
calculated in index form to match Vq. PMTq
will be constant throughout the stress test for fixed-rate loans
because rc,q is fixed at rc,0.
c. Calculate a remaining loan balance index for the UPB
outstanding at the beginning of each quarter of the stress period,
UPBq, based on PMTq, Tr, and
elapsed time in the stress period, q, using the formula:
[GRAPHIC] [TIFF OMITTED] TP13AP99.071
[GRAPHIC] [TIFF OMITTED] TP13AP99.297
3. Compute updated LTV ratios (LTVq) for each quarter
of the stress period:
[GRAPHIC] [TIFF OMITTED] TP13AP99.072
4. Compute the standard deviation of house price growth paths
(d,q) around the HPId,q value. Limit
the value of the age variable to avoid negative ``diffusion.''
Negative diffusion occurs when the variance of house prices declines
over time. The quadratic formula used here for the standard
deviation of individual house price index values will create
negative diffusion unless age is limited. The age limit formula is
found by solving the first derivative of the house price volatility
variance with respect to age, for zero. This variance is the
function under the root sign in the d,q equation
below (but using Aq rather than MAq). The age
limit gives the value of age for which the diffusion of house price
growth is maximized. Once this age value is reached, the stress test
then holds diffusion at the maximum value for the remainder of the
life of the loan:
[GRAPHIC] [TIFF OMITTED] TP13AP99.073
[[Page 18246]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.298
5. Calculate the probability of negative equity in each stress
period quarter:
[GRAPHIC] [TIFF OMITTED] TP13AP99.074
[GRAPHIC] [TIFF OMITTED] TP13AP99.299
[GRAPHIC] [TIFF OMITTED] TP13AP99.373
[[Page 18247]]
3.5.2.3.2.4 Burnout (Bq)
[a] The prepayment ``burnout'' variable, Bq,
indicates whether there have been at least two quarters of
``significant refinance opportunities'' among the previous eight
quarters of loan life. A mortgage undergoes a significant refinance
opportunity when its coupon is at least two percentage points above
the then-prevailing rate on 30-year mortgages. The rate on 30-year
mortgages is always used as the benchmark for defining refinance
opportunities, regardless of the type of mortgages being analyzed.
Prepayment burnout is a binary variable--two quarters of significant
refinance opportunities either occur or do not occur.
[b] If Aq 8, then Bq=0. If
Aq >8, then:
[GRAPHIC] [TIFF OMITTED] TP13AP99.075
[GRAPHIC] [TIFF OMITTED] TP13AP99.300
3.5.2.3.2.5 Occupancy Status (OS)
The occupancy status variable is the percentage of loans in an
Enterprise portfolio that are investor-owned (rental) properties
rather than owner-occupied properties. It is a constant value (OS)
applied equally to all loan groups and in all stress period
quarters, computed as follows:
[GRAPHIC] [TIFF OMITTED] TP13AP99.076
[GRAPHIC] [TIFF OMITTED] TP13AP99.301
3.5.2.3.2.6 Relative Spread (RSq)
The relative spread variable (RSq in the formula
below) is the percentage spread between a loan's contract rate and
the rate on 30-year fixed-rate mortgages in the current quarter of
the stress test. The higher this percentage is, the more likely a
loan is to prepay:
[GRAPHIC] [TIFF OMITTED] TP13AP99.077
3.5.2.3.2.7 Yield Curve Slope (YSq)
The variable YS q in the formula below represents the
slope of the yield curve. It is included in the prepayment
calculations to represent different relationships between short-and
long-term interest rates. Different yield curve slopes represent
different relationships between short and long term interest rates,
and these relationships impact incentives to refinance either into
ARMs or into fixed-rate mortgages:
[GRAPHIC] [TIFF OMITTED] TP13AP99.078
[GRAPHIC] [TIFF OMITTED] TP13AP99.302
[[Page 18248]]
3.5.2.3.2.8 Product Type Adjustment Factors
Product types other than fixed-rate 30-year mortgages and ARMs
receive unique product-specific adjustment factor weights in the
stress test. These factors relate the default and prepayment risk of
each product type to the fixed-rate 30-year mortgage. ARMs do not
need a risk adjustment factor because they use separate default and
prepayment equations. All products other than 30-year fixed-rate and
adjustable-rate mortgages use the same pair of default and
prepayment equations. The product types included in this combination
grouping, which receive product-specific risk adjustment factors,
are: 20-year fixed-rate, 15-year fixed-rate, balloon, government
insured or guaranteed loans, and second mortgages. All loan products
with payment changes, such as graduated payment mortgages, two-step
mortgages, and buydown mortgages, are treated as ARMs and use the
ARMs default and prepayment formulas without a product adjustment
factor. Biweekly and reverse mortgages are included with standard
monthly mortgages of similar term and do not therefore require
separate adjustments. The adjustment factor values are provided in
Table 3-17.
3.5.2.3.2.9 Benchmark Calibration Factor
A calibration adjustment of 0.146 is added to each statistical
default equation to reasonably relate current loan default rates to
the historical benchmark experience. The value 0.146 is a weighting
factor, not an explanatory variable.
3.5.2.3.3 Combining Explanatory Variables and Weights
[a] Each explanatory variable outlined above has associated
numerical weights that are used in default and prepayment rate
calculations. These weights, which are the estimated coefficients
from statistical regressions, are referred to here as beta factors,
j, for default weights, and gamma factors,
k, for prepayment weights. As mentioned above,
there is also a constant weight for benchmark calibration. In
addition, each statistical equation has a different regression
constant. These constants appear as separate weights, not tied to
any explanatory variables.
[b] The weights are combined to compute two sums:
Xq for defaults and Xq for
prepayment as follows:
[GRAPHIC] [TIFF OMITTED] TP13AP99.079
[GRAPHIC] [TIFF OMITTED] TP13AP99.303
[c] The only explanatory variables for which both the variable
and its weight are included in the formula above are age
(Aq), age squared (A2q), occupancy status (OS)
and burnout (Bq). For each of these variables, the
variable value is multiplied by its weight, which can be found in
Table 3-17. For other (categorical) explanatory variables, however,
the weights are not accompanied by the actual values of the
explanatory variables. For these variables the computed variable
value is only used to identify the category to which it belongs so
that a representative weight can be selected from the weight table
(Table 3-17) of this Appendix. Only the obtained weight is included
in the Xq and Xq formulas
for these variables.
[[Page 18249]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.268
[[Page 18250]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.269
3.5.2.3.4 Calculating Default and Prepayment Rates
The total weighting factors, Xq and
Xq, are converted into quarterly default and
prepayment probabilities using the following logistic probability
equations:
[GRAPHIC] [TIFF OMITTED] TP13AP99.080
[GRAPHIC] [TIFF OMITTED] TP13AP99.081
[GRAPHIC] [TIFF OMITTED] TP13AP99.304
3.5.2.3.5 Monthly Default and Prepayment Rates
To this point, all calculations involved creating quarterly time
series of values throughout the ten-year stress period (40
quarters). In this step, the quarterly conditional default and
prepayment rates are converted into monthly rates as follows:
[[Page 18251]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.082
[GRAPHIC] [TIFF OMITTED] TP13AP99.083
[GRAPHIC] [TIFF OMITTED] TP13AP99.305
3.5.2.4 Output
Use the resulting 120 monthly conditional default and prepayment
rates for each loan group to calculate monthly principal reductions
resulting from defaults and prepayments, and to calculate default
losses for each month in the ten-year stress period.
3.5.3 Single Family Loss Severity
3.5.3.1 Overview
[a] The Single Family Loss Severity component of the stress test
computes loss severity rates for single family mortgages that
default in each month of the stress test. The loss severity rate is
the net cost of a loan default expressed as a percentage of the
unpaid principal balance (UPB) at the time of default. Based on
various cost and revenue elements associated with a loan default,
the stress test calculates loss severity rates as the present value
(at default date) of the net cash flows that occur following the
default date. Most cost and revenue elements are entered as constant
rates across loan groups throughout the stress period. Two
exceptions are proceeds from property disposition and asset funding
costs. Proceeds are derived through a formula that uses both
historical and stress period house price appreciation rates, and
that accounts for loan amortization from origination through
default. Funding cost of the defaulted mortgages and the resulting
foreclosed properties is captured by discounting the loss severity
elements, using a cost-of-funds interest rate that varies during the
stress period. Loss severity rates throughout the stress period will
also vary according to the application of percent-denominated credit
enhancements (dollar-denominated credit enhancements are directly
applied in the Cash Flow component of the stress test) and their
associated credit ratings.
[b] The inputs used to compute loss severity rates include
several starting position loan group characteristics, counterparty
credit risk factors, historical house price index series and stress
period house price growth rates, house price appreciation volatility
parameters, and stress test interest rate series. The output of loss
severity rates for each loan group are used in the Cash Flow
component of the stress test (see section 3.9, Cash Flows, of this
Appendix) to calculate (dollar) default losses.
3.5.3.2 Inputs
[a] The Single Family Loss Severity component of the stress test
uses loan group characteristics as of the start of the stress test,
including information on certain types of credit enhancements, and
credit risk factors associated with counterparty rating categories
(see section 3.6, Other Credit Factors, of this Appendix). In
addition, it uses historical and stress period HPI series, house
price appreciation volatility parameters, and one interest rate
series (see section 3.4, Property Valuation, of this Appendix).
[b] The particular loan group characteristics (refer to section
3.1, Enterprise Data, of this Appendix for the definitions of these
loan group characteristics), with associated variable names used in
the procedures below, are:
Product Type
Portfolio (retained or sold portfolio)
Origination Year (subscript ``y'')
Origination Month (tm, for commitment loan
groups only)
Census Division (subscript ``d'')
Starting Coupon (rc,s)
Original Coupon (rc,0, only used for ARMs)
Passthrough Rate (rp, for sold loans only)
Original LTV (LTV0)
Mortgage Age (As)
Amortization Term (Ta)
Credit Enhancement Coverage Type 1 (Cmi, PMI
coverage rate)
Credit Enhancement Coverage Type 2 (Crc,
seller/servicer recourse coverage rate)
Percent of UPB under ``AAA'' coverage in a loan group
(CR)
Percent of UPB under ``AA'' coverage in a loan group
(CR)
Percent of UPB under ``A'' coverage in a loan group
(CR)
Percent of UPB under ``BBB'' coverage in a loan group
(CR)
[c] Credit enhancement coverages, both Type 1 and Type 2, are
reduced throughout the stress test according to ``haircuts,'' as
defined in section 3.6, Other Credit Factors, of this Appendix.
These haircuts represent percentage reductions to credit enhancement
coverage due to the inability of a counterparty to meet its
obligations under stressful conditions. The final (end-of-stress-
period) haircuts, by credit rating class (AAA, AA, A, and BBB), are
obtained from section 3.6, Other Credit Factors, of this Appendix.
[d] In addition, historical Census division HPI series and house
price appreciation volatility parameters are obtained from the most
recently available OFHEO HPI Report. The HPI series are used to
update collateral property values to the beginning of the stress
test. Property values are then updated during the stress period with
monthly house price growth rates obtained from section 3.4, Property
Valuation, of this Appendix. The historical volatility parameters
are used with stress period property values to develop distributions
of property values and levels of home equity within loan groups.
[e] The final input used here is the six-month Federal agency
cost-of-funds rate, for each month of the stress period (variable
``rd,t''). This monthly series is generated by the
interest rate component of the stress test (See section 3.3,
Interest Rates, of this Appendix) and is used as the discount rate
for computing the present value of the three major elements of the
loss severity rate--defaulting UPB, net costs or proceeds associated
with foreclosure, and net cash flows from holding and disposition of
Real Estate Owned (REO) property.
3.5.3.3 Procedures
[a] The process of deriving loss severity rates involves
calculating the present value of three loss elements. The first loss
element (PV1) is the amount of defaulting UPB. The second loss
element (PV2) is the expense related to foreclosure, net of any
mortgage insurance proceeds. The third loss element (PV3) combines
post-foreclosure property expenses with proceeds from REO property
disposition. Each of these three loss elements is computed as the
present value (as of the default date) of the net cash flows
occurring at a separate point in time--four months after default for
the first loss element, 13 months after default for the second loss
element, and 20 months after default for the third loss element. The
present values of the three loss elements then are added together to
derive an initial loss severity rate (NPV1). Finally, available
seller/servicer recourse against the (initial) loss is applied to
calculate the final loss severity rate (NPV3). Figure 3-3 of this
Appendix depicts the timing of the three loss elements and how they
are combined to produced initial and final loss severity rates.
[b] In the procedures for calculating loss severity rates, loan
amortization is performed each month for surviving loans in each
loan group; all discounting of cash flows uses semi-annual
compounding of interest rates; all calculations add expenses and
subtract revenues to calculate loss severity rates; and all loss
elements are calculated as percentages of the UPB of the defaulting
loans. With the exception of computations for FHA and VA loans,
calculations are not specific to any particular loan product types,
although loan group characteristics (coupon rate and amortization
term) are used in the severity calculations.
[c] The lack of product type distinctions in severity
calculation means that adjustable
[[Page 18252]]
rate mortgages are treated like fixed-rate mortgages. Their coupon
rates are not updated during the stress test, and the original
coupon is used to perform loan amortizations used in the statistical
equation for property disposition proceeds. This simplification does
not affect the actual defaulting UPB used to calculate dollar
losses. The cash flow portion of the stress test does update coupon
rates for adjustable rate products, and uses the updated rates to
amortize loan group UPB. There are also no differences in loss
severity rate calculations for investor loans. The stress test does
not group loans according to occupancy status (owner-occupant versus
investor/rental), although the statistical analysis used to derive
the loss severity elements for the stress test used data on both
occupancy status types. Thus, the loss severity elements shown here
reflect a balance of owner-occupant and investor loans.
[d] The stress test groups FHA and VA loans together. To
calculate severity rates, FHA and VA insurance coverage amounts are
calculated separately for all FHA/VA loan groups. Loan group credit
enhancements are then calculated by summing the coverage amounts,
with FHA insurance receiving a 0.67 weight and VA insurance
receiving a 0.33 weight. Final loss severity rates for FHA/VA loan
groups are then computed based on these weighted average coverage
amounts.
3.5.3.3.1 Defaulting UPB
The defaulting UPB is the first loss element included in the
loss severity rate calculation. The stress test recognizes
defaulting UPB four months after the month of default. At this
point, the defaulting UPB is recognized as a loss severity element
and a potential cost (pending offsetting revenues from mortgage
insurance and property disposition). For sold loans, defaulting
mortgages are first purchased from the security pools, requiring a
cash outlay equal to the UPB. Because only sold loans involve actual
cash outlays, sold and retained loans are treated slightly
differently in this loss element calculation.
[GRAPHIC] [TIFF OMITTED] TP13AP99.374
1. For sold loans, recognize the cash outlay by discounting UPB
back to the date of default:
[GRAPHIC] [TIFF OMITTED] TP13AP99.084
[GRAPHIC] [TIFF OMITTED] TP13AP99.385
2. For retained loans, set PV1t = 1 to represent the
full UPB. No discounting is necessary because recognition of the
defaulting UPB does not involve an actual cash outlay.
3.5.3.3.2 Net Costs or Proceeds Associated with Foreclosure
The second loss element includes foreclosure related
transactions. There are
[[Page 18253]]
several cash flows, so that multiple computations are required.
1. Calculate survival factors for each counterparty rating
category, for each month of the stress period. The monthly survival
factors represent percentages of obligations that counterparties
with given credit ratings are expected to meet as the stress period
continues. They are derived from the final haircuts defined in
section 3.6, Other Credit Factors, of this Appendix. These factors
are applied here to private mortgage insurance (PMI) coverage, and
later to seller/servicer recourse obligations. Survival factors for
each credit rating category are constant across loan groups:
[GRAPHIC] [TIFF OMITTED] TP13AP99.085
[GRAPHIC] [TIFF OMITTED] TP13AP99.306
2. Calculate private mortgage insurance (PMI) proceeds.
a. Calculate the weighted average survival factor for each loan
group. For each month, t, of the stress period, multiply the
survival factor for each counterparty rating, SFR,t, by
the percentage of the loan group UPB covered by counterparties with
the same rating, CR. Sum the results across all
counterparty ratings, R. Next, divide that sum by the sum of all
counterparty coverage percentages. This produces a weighted average
survival factor, SFw,t, by loan group, for each month, t,
of the stress period:
[GRAPHIC] [TIFF OMITTED] TP13AP99.086
[GRAPHIC] [TIFF OMITTED] TP13AP99.307
b. Multiply the weighted average survival factors,
SFw,t, by the PMI percentage coverage rate,
Cmi, to derive monthly adjusted percentage coverage
rates, Cmi,t:
[GRAPHIC] [TIFF OMITTED] TP13AP99.087
[GRAPHIC] [TIFF OMITTED] TP13AP99.308
c. Compute mortgage insurance proceeds (mit), by
multiplying the adjusted PMI percentage coverage rate,
Cmi,t, by the mortgage insurance claim amount. First, for
all conventional loans--loan groups other than FHA/VA:
[GRAPHIC] [TIFF OMITTED] TP13AP99.088
For FHA/VA loan groups, calculate the FHA insurance proceeds:
[GRAPHIC] [TIFF OMITTED] TP13AP99.089
[[Page 18254]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.309
3. Discount all foreclosure related cash flows by
tf=13 months to compute the post-foreclosure loss
element, PV2t.
a. For retained loans:
[GRAPHIC] [TIFF OMITTED] TP13AP99.090
b. For sold loans, add passthrough interest expense to mortgage-
backed security holders for 4 months:
[GRAPHIC] [TIFF OMITTED] TP13AP99.091
c. For FHA/VA loans:
[GRAPHIC] [TIFF OMITTED] TP13AP99.092
[GRAPHIC] [TIFF OMITTED] TP13AP99.310
4. Calculate the payment to the loan servicer (PVSt)
net of any interest paid by the seller/servicer to the Enterprise
that would be repaid in the post-foreclosure servicer claim. The
present value factor generated here is not used in the computation
of the foreclosure loss component, but will be used later to account
for cases where there is full recourse to the seller/servicer. This
is required only where there is Type 2 Credit Enhancement coverage.
It is not used for FHA/VA loans. For retained loans:
[GRAPHIC] [TIFF OMITTED] TP13AP99.093
For sold loans, add the (4 months) interest passed through by
the Enterprise to security holders:
[[Page 18255]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.094
[GRAPHIC] [TIFF OMITTED] TP13AP99.311
3.5.3.3.3 Net Cash Flow from Holding and Disposition of REO Property
The third loss element includes cash flows associated with
management and disposition of REO property. Cash flows used in
calculating this element are sales proceeds from disposition of
foreclosed property and REO property management (maintenance and
operating) expenses.
3.5.3.3.3.1 Calculate Proceeds From Property Sale
Sales proceeds is a dynamic loss severity element whose
calculation involves updating property values and loan balances over
time. Several steps are required. First, property values and UPB are
updated from origination to the time of default. This is done with
index values, rather than dollar values. Property values are
represented by a house price index, and loan balances by a UPB index
(ratios of defaulting UPB to the original house price). Second, a
statistical measure (z-score) of the distance between the logarithm
of house price index and the logarithm of the loan balance index is
calculated. Third, an econometric equation uses the z-score to
compute the portion of UPB that is not recovered at property
disposition. Finally, the unrecovered portion of UPB is converted
into proceeds from property sale.
1. Update property values.
a. Calculate a house price index at the start of the stress
test, according to origination year and Census division cohort:
[GRAPHIC] [TIFF OMITTED] TP13AP99.095
[GRAPHIC] [TIFF OMITTED] TP13AP99.312
Because HPI values are as of the end of each quarter,
HPId,-1 gives the value as of the start of the stress
period. The OFHEO HPI is published beginning with the first quarter
of 1980. OFHEO has also produced (but not published) values for
earlier years. To season loans originating in 1979, assign
HPId,q according to the Census division specific values
listed in Table 3-16. Treat all pre-1979 originations as if they
were originated in 1979.
b. Calculate house price index values during the stress period
by multiplying the Id,q by cumulative house price growth
rates in the stress period:
[GRAPHIC] [TIFF OMITTED] TP13AP99.096
[GRAPHIC] [TIFF OMITTED] TP13AP99.313
[[Page 18256]]
Do not calculate Id,q for loans that an Enterprise
has committed to buy, but not yet purchased at the beginning of the
stress period, because pre-stress period house price appreciation is
not applicable. The house price index for these loans is the
cumulative monthly growth rate from the month after delivery to the
month of loss severity calculations (month of default):
[GRAPHIC] [TIFF OMITTED] TP13AP99.097
[GRAPHIC] [TIFF OMITTED] TP13AP99.314
2. Calculate the standard deviation of house price growth paths,
d,t, around the average growth path implied by
the HPId,q,t value. This first requires limiting the
value of the age variable to avoid negative ``diffusion.'' Negative
diffusion occurs when the variance of house prices declines over
time. While negative diffusion is not expected to happen in
practice, the formula for the standard deviation of house price
growth paths (which is a quadratic function of time, where the
first-order term is positive and the second-order term is negative)
will create negative diffusion unless age is limited.
a. Create a variable for mortgage age in the stress test:
[GRAPHIC] [TIFF OMITTED] TP13AP99.098
[GRAPHIC] [TIFF OMITTED] TP13AP99.315
b. Create a mortgage age variable (MAt) that limits
the mortgage age to a maximum value:
[GRAPHIC] [TIFF OMITTED] TP13AP99.099
[GRAPHIC] [TIFF OMITTED] TP13AP99.316
c. Calculate the standard deviation of house price growth rate
path using MAKt:
[GRAPHIC] [TIFF OMITTED] TP13AP99.100
[GRAPHIC] [TIFF OMITTED] TP13AP99.317
3. Compute a monthly loan payment factor using the original
coupon rate and original LTV (LTV0). Since original
property value is specified to be equal to one, LTV0
represents the original UPB. Use this payment factor to compute the
time series of UPB index (see below) to capture amortization of
surviving loans in each loan group throughout the stress period:
[GRAPHIC] [TIFF OMITTED] TP13AP99.101
[[Page 18257]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.318
4. Calculate the time series of UPB index--the ratios of
defaulting UPB in each month of the stress period to the original
house price:
[GRAPHIC] [TIFF OMITTED] TP13AP99.102
[GRAPHIC] [TIFF OMITTED] TP13AP99.319
5. Compute the z-score for the ``distance'' between the logarithm
of the house price index and the logarithm of the UPB index. The use of
logarithmic values allows each variable to be specified as a percentage
difference from the original property value (1.0). This transformation
makes the distance between the house price and UPB indexes consistent
with the standard deviation of the house price growth rates used to
calculate the z-score.\11\ The formula for the z-score is:
---------------------------------------------------------------------------
\11\ This standard deviation is of cumulative house price growth
rates. The log of HPI is the cumulative growth of average house
prices in the geographic area, while the log of b gives an HPI-
growth-rate-equivalent interpretation to owner invested equity
(downpayment plus amortization). The resulting log difference is the
amount by which the individual house price growth must be lower than
average market growth in order to eliminate any equity in the
property and thus lead the borrower to consider default.
[GRAPHIC] [TIFF OMITTED] TP13AP99.103
[GRAPHIC] [TIFF OMITTED] TP13AP99.320
The allowable values of zt are bounded by 4.0 and -
0.50. If the computed value zt is outside either of these
bounds, it is reset to its closest boundary value.
6. Compute the percentage of UPB that is not recovered at
property disposition based on the statistically derived relationship
between the percentage of UPB unrecovered at property disposition
and the z-score:
[GRAPHIC] [TIFF OMITTED] TP13AP99.104
[[Page 18258]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.321
Because log-transformed values of the unrecovered UPB
(ln(Lt) + 1)) were used in the regression, the ``1'' in
the equation above is a result of using the antilog to derive the
formula for Lt. In addition, the formula also includes
the calibration factor to reasonably relate loss severity rate to
the benchmark experience.
7. Calculate sales proceeds from the disposition of each
foreclosed property, Pt, as UPB less the portion that was
not recovered at disposition, Lt:
[GRAPHIC] [TIFF OMITTED] TP13AP99.105
[GRAPHIC] [TIFF OMITTED] TP13AP99.322
3.5.3.3.3.2 Net Cash Flow at Property Disposition
Subtract sales proceeds from expenses related to REO property,
then discount the result by (tf+ti = 20
months) to obtain the present value of the third loss severity
element:
[GRAPHIC] [TIFF OMITTED] TP13AP99.106
[GRAPHIC] [TIFF OMITTED] TP13AP99.107
or
[GRAPHIC] [TIFF OMITTED] TP13AP99.323
3.5.3.3.4 Final Calculations of Loss Severity Rates
At this point, all cost elements of loss severity are included
in PV1, PV2, and PV3. Revenues from private mortgage insurance (Type
1 credit enhancement) or FHA insurance are also included in PV2. The
sum of PV1, PV2, and PV3 then provides an initial net-present-value
loss severity rate (NPV1). Once this is calculated, potential
revenues from seller/servicer recourse (Type 2 credit enhancement)
and VA insurance guaranty proceeds are computed. For non-government
(conventional loans), the recourse proceeds are subtracted from NPV1
to arrive at final loss severity rates (NPV3) for each loan group,
in each month of the stress test. For FHA/VA loan groups, final loss
severity rates are calculated using a weighted average of the
proceeds from the two forms of government insurance.
1. Calculate the initial loss rates (after mortgage insurance
and FHA coverage, but before seller/servicer recourse or VA
coverage):
[GRAPHIC] [TIFF OMITTED] TP13AP99.376
[[Page 18259]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.324
2. Proceed based upon whether the loan group represents
conventional or FHA/VA loans:
a. For conventional loans, check the initial losses in
NPV1t to evaluate whether there is any loss remaining.
Loans with losses less than zero, where NPV1t
0, will not receive any additional credit for seller/servicer
recourse. For those loans, set RCtt = 0, and proceed to
Step 6. Otherwise, if NPV1t > 0, go to Step 3.
b. For FHA/VA loans, proceed to Step 5.
3. Re-calculate initial loss severity rates using the full
seller/servicer claim amount, PVSt, rather than the post-
insurance foreclosure cash flow, PV2t:
[GRAPHIC] [TIFF OMITTED] TP13AP99.108
[GRAPHIC] [TIFF OMITTED] TP13AP99.325
4. Use NPV2t with appropriate percentage recourse
(Type 2) coverage rates and survival factors to calculate seller/
servicer recourse coverage amounts, RCt:
[GRAPHIC] [TIFF OMITTED] TP13AP99.109
[GRAPHIC] [TIFF OMITTED] TP13AP99.326
Go to Step 6.
5. For FHA/VA loan groups, calculate the effective loss rate after
recourse coverage amounts provided by VA guarantees:
[GRAPHIC] [TIFF OMITTED] TP13AP99.110
and then:
[GRAPHIC] [TIFF OMITTED] TP13AP99.377
[[Page 18260]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.327
6. Calculate final loss severity rates separately for
conventional and FHA/VA loans.
a. For conventional loans, subtract the percent reduction in net
losses provided by recourse coverage, RCt, from the
initial loss severity rate, NPV1t:
[GRAPHIC] [TIFF OMITTED] TP13AP99.111
[GRAPHIC] [TIFF OMITTED] TP13AP99.328
b. For FHA/VA loans, compute a weighted-average loss severity
rate, using NPV1t--which includes FHA insurance--and
NPVVAt--which includes VA insurance:
[GRAPHIC] [TIFF OMITTED] TP13AP99.378
[GRAPHIC] [TIFF OMITTED] TP13AP99.329
3.5.3.4 Output
The resulting 120 monthly loss severity rates (NPV3t)
for each loan group are used as inputs by the Cash Flow component of
the stress test to calculate monthly (dollar) default losses (see
section 3.9, Cash Flows, of this Appendix).
3.5.4 Multifamily Default and Prepayment
3.5.4.1 Overview
The Multifamily Default and Prepayment component of the stress
test calculates the monthly rates of default and prepayment for each
multifamily loan group throughout the stress period. The process of
computing default and prepayment rates requires input data on: one,
loan group characteristics in the Enterprise portfolios at the
beginning of the stress test; two, historical rent growth rate
information to update values of collateral for the loans to the
beginning of the stress test; and three the economic conditions of
the stress period--interest rates, vacancy rates and rent growth
rates. These input data are used to create values for the
explanatory variables in the Multifamily Default and Prepayment
component. The annual-equivalent default and prepayment rates for
each month of the stress period are generated using the values of
the explanatory variables and the regression coefficients (or
weighting factors). These coefficients are based on statistical
analysis of the relationship between default and prepayment rates
and the explanatory variables. Finally, the annual-equivalent rates
are converted into monthly rates for use in the Cash Flow component
of the stress test to simulate loan terminations and associated
credit losses.
3.5.4.2 Inputs
Inputs for the Multifamily Default and Prepayment component of
the stress test include loan group characteristics, interest rate
series, historical rent indexes, and stress period rent growth rates
and vacancy rates. Each of these are discussed below.
3.5.4.2.1 Loan Group Characteristics
As described in section 3.1, Enterprise Data, of this Appendix,
multifamily loan group characteristics data are generated through
aggregation of individual Enterprise loans as of the beginning of
the stress test, according to defined aggregation criteria. The
characteristics of a loan group include both categorical and
continuous variables. The values of categorical variables indicate
the range within which a loan group characteristic falls (``value-
range''). The values of continuous variables are averages of the
values of the characteristics of the underlying loans, where the
weights are the unpaid principal balances of each loan in the group,
at the start of the stress test. The following are loan group
characteristics used in the Multifamily Default and Prepayment
component of the stress test (using allowable values for each
variable found in section 3.1, Enterprise Data, of this Appendix):
Origination year
Census region
Metropolitan statistical area
Product type
Mortgage program
Original LTV (the variable ``LTV0'' in
equations below)
Original coupon (the variable ``rc,0'' in
equations below)
Starting coupon (the variable ``rc,s'' in
equations below)
Starting UPB (the variable ``UPBs'' in
equations below)
Debt coverage ratio, at the time of purchase by the
Enterprises (the variable ``DCR0'' in equations below)
Amortization term, in months (the variable
``Ta'' in equations below)
Mortgage age, in months (at the start of the stress
period; the variable ``As'' in equations below)
3.5.4.2.2 Interest Rate Series
Three interest rate series are used in the Multifamily Default
and Prepayment component. These series are generated from the
Interest Rate component of the stress test, for each month of the
stress period (see section 3.3, Interest Rates, of this Appendix).
Historical values for one of these interest rate series are also
required. (See below. Note that all interest rate series are in
decimal format.) The particular input series are:
Federal Home Loan Bank 11th District Cost of Funds
Index (COFI) (the variable ``rb,t'' in equations below)
Conventional mortgage rate (the variable
``rf'' in equations below)
Ten-year Constant Maturity Treasury Yield (the variable
``y120t'' in equations below). Historical values of this
variable are input starting 30 years prior to the start of the
[[Page 18261]]
stress test, and stress test simulation values are used to extend
the series throughout the stress period.
3.5.4.2.3 Historical Rent Indexes
Updating property values of collateral for multifamily loans at
the beginning of the stress test requires use of rent indexes. The
stress test uses the residential rent component of the Consumer
Price Index (CPI), which is available from the U.S. Department of
Labor, Bureau of Labor Statistics (BLS). The series required for
this part of the stress test are those for the U.S., the four Census
regions, and the 29 Metropolitan Statistical Areas (MSAs) covered by
the BLS surveys.
3.5.4.2.4 Stress Period Vacancy Rates and Rent Growth Rates
Monthly vacancy rate and rent growth rate series for the stress
period are generated by the Property Valuation component of the
stress test (see section 3.4, Property Valuation, of this Appendix).
These series are used to update multifamily property values
throughout the stress period.
3.5.4.3 Procedures
[a] Separate default equations are used to distinguish between
loans acquired through: one, cash purchases and two, negotiated
transaction. In a cash purchase, an Enterprise acquires a newly
originated loan that meets standard underwriting guidelines; the
purchase can include recourse to the seller/servicer. In a
negotiated transaction, an Enterprise generally acquires a pool of
seasoned, nonconforming loans.
[b] FHA-insured loans are a subset of loans that are purchased
through negotiated transactions, but they are included with the cash
transaction loans for default calculation purposes.
[c] Fixed-rate multifamily loans have prepayment restrictions,
for example, yield maintenance fees and lockouts, that severely
limit prepayments for about two-thirds of the loan term. To account
for the differences in prepayment speeds that result from these
restrictions, five prepayment equations are used for the following
types of loans: fixed-rate loans in the restriction period, fixed-
rate balloon loans beyond the restriction period, self-amortizing
fixed-rate loans beyond the restriction period, balloon loans at the
balloon point, and adjustable rate mortgages.
[d] To calculate default and prepayment rates in the stress
text, the input data described above are used to compute the values
of explanatory variables for the equations for multifamily default
and prepayment rates. A total of 16 explanatory variables (shown in
Table 3-20) are computed for each loan group, and for each month of
the stress period. The following describes calculations of
explanatory variables and the resulting default and prepayment
rates. Unless otherwise indicated, each variable subscripted with a
``t'' is computed for the 120 months of the stress period. To
illustrate each procedure, formulas are shown for one loan group for
each month of the stress test. The same logic applies to all loan
groups.
[e] The values of explanatory variables in each month are used
in the default and prepayment equations to calculate annual default
and prepayment rates. The stress test computes default and
prepayment rates that would result if the conditions prevailing in
each month were to continue for an entire year. These annual rates
are converted to monthly rates for use in section 3.9, Cash Flows,
of this Appendix.
3.5.4.3.1 Computation of Explanatory Variables
3.5.4.3.1.1 Mortgage Age (At, AYt)
[a] Mortgage age in each month of the stress period is
calculated as:
[GRAPHIC] [TIFF OMITTED] TP13AP99.113
[GRAPHIC] [TIFF OMITTED] TP13AP99.330
[b] Since mortgage age enters the default and prepayment
equations in years, rather than in months, an age-in-years variable,
AYt, is created:
[GRAPHIC] [TIFF OMITTED] TP13AP99.114
3.5.4.3.1.2 Program Restructuring (PR)
The stress test differentiates between cash programs in effect
before 1988 for Fannie Mae and before 1992 for Freddie Mac
(``original programs'') and later cash programs. This
differentiation accounts for the greater credit risk of the earlier
cash programs. The variable PR is used in two ways to adjust
original program loan groups for this greater risk. PR is only used
for loans in the cash programs (except FHA-insured loans) because
OFHEO has identified the program structure deficiencies that caused
this greater risk only on these loans. The variable is not used to
adjust the risk profile of loans acquired through negotiated
programs. The PR variable is computed for each loan group according
to the following:
[GRAPHIC] [TIFF OMITTED] TP13AP99.115
First, PR is used as a categorical variable to distinguish the
original cash programs from more recent cash programs of the
Enterprises (``current programs''). This usage of PR captures the
higher default risk of the Enterprises' original programs. Second,
PR is used as a flag for when to adjust DCR0 and
LTV0 for overly optimistic appraisal practices inherent
in original cash program loans. (See sections 3.5.4.3.3.10, Formula
for Constructing the DCR Time Series and 3.5.4.3.4.4, Construct the
LTV Time Series, of this Appendix.)
3.5.4.3.1.3 Value of Depreciation Write-off (DW)
The present value of tax benefits afforded to an investor/owner
in a multifamily property is captured in a depreciation write-off
variable (DW). Based on depreciation rules and OFHEO's estimates of
the marginal tax rate for ordinary income, the marginal tax rate for
capital gains, and the risk-adjusted return for multifamily
projects, a value of 9.27 for this variable (DW) is used in the
stress test. This value represents a 9.27 percent estimated return
for a 20-year holding period on investments in multifamily property
resulting from tax benefits associated with ownership and taxes paid
on the ultimate sale of the property, based on 1995 data. OFHEO may
change the value for this variable if there are significant changes
in depreciation rules or tax rates. DW affects defaults and is held
constant for
[[Page 18262]]
all cash programs throughout the stress period. However, it is not
used to project default rates of negotiated programs.
3.5.4.3.1.4 Seller/Servicer Repurchase Flags (RF, RA)
[a] Mortgage default in the stress test is defined as a loan
termination in which the borrower must relinquish title to the
property because of an inability to make loan payments. However,
there is one exception for multifamily mortgages in certain
negotiated programs. In these negotiated programs, when a loan
becomes 90 days delinquent, the seller/servicer must buy the loan
out of the pool and attempt to resolve the delinquency. For these
loans, the stress test defines default as a 90-day delinquency,
rather than a full default. The occurrence of 90-day delinquencies
is always higher than the occurrence of full defaults, since many
90-day delinquent loans cure or are modified.
[b] To distinguish a ``90-day delinquency'' type of default from
a full default, the stress test includes two categorical variables
that flag fixed-rate (RF) and adjustable rate (RA) negotiated
program loans with repurchase requirements:
[GRAPHIC] [TIFF OMITTED] TP13AP99.116
3.5.4.3.1.5 Joint Probability of Negative Equity and Negative Cash
Flow (JPt)
The joint probability of negative equity and negative cash flow
(JPt) is defined as the probability that any given loan
will simultaneously experience a loan-to-value ratio
(LTVt) greater than 1.00 and a debt coverage ratio
(DCRt) less than 1.00. JPt is the principal
variable used in the stress test to measure the value of default to
multifamily borrowers. Creating this variable involves updating
DCRt and LTVt over time using a property net
operating income (NOI) growth factor, changes in mortgage payments,
loan amortization, and a capitalization rate multiplier. The NOI
growth factor is updated over time using vacancy rate changes and
rental inflation since loan origination. The capitalization rate
multiplier is updated based on changes in interest rates since loan
origination.
3.5.4.3.2 Updating Average Property Income
3.5.4.3.2.1 Create Rent Indexes for the Start of the Stress Period
Rent indexes at the start of the stress period are created using
time series of annual percent changes in the residential rent
component of the CPI for each of the four Census regions and the 29
MSAs covered by BLS surveys. If the stress test begins at a time
other than January 1 (first quarter of the year), the residential
rent component of the CPI at the end of the quarter just preceding
the start of the stress test is used to create the final ``year'' of
the rent index time series. Most MSA level CPI series produced by
BLS start in 1970, but some do not begin until the 1980s. The
regional CPI series are available beginning in 1978, so percent
changes for these can only be computed starting in 1979. Each
regional and MSA percent-change series is constructed as follows:
1. Fill-in the pre-1979 regional series with percent changes in
the rent index values for the national CPI, going back 30 years from
the start of the stress test. If any MSA is missing one or more
years of data, fill-in missing values from regional series. This
results in 33 time series of annual rent growth rates for 30 years,
ending in the year and quarter just preceding the beginning of the
stress test.
2. Using these time series, create the rent index value for each
loan group at the start of the stress period, as a cumulative index
from the loan origination year to the start of the stress test:
[GRAPHIC] [TIFF OMITTED] TP13AP99.117
[[Page 18263]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.331
3. In order to link the rental series to loan group
characteristics, first match each loan group by MSA code to the
available residential rent series from BLS. If there is a match,
then use that MSA series of historical annual growth rates of
residential rent, as described above, to generate the value for
Im,y. If the loan group is not in an MSA covered by the
BLS residential rent series, then match the Census region of the
property to the appropriate regional residential rent series, and
use the regional historical annual growth rates of the residential
rent series to generate the value for Im,y. Assume that
all loans originate in the middle of the year, for purposes of the
first-year rent growth rate. To accomplish this, the above formula
uses the square root of the growth rate in the year of loan
origination.
3.5.4.3.2.2 Update Each Rent Index throughout Stress Period
The rent index at the beginning of the stress test
(Im,y) is updated, for each loan group, throughout the
stress period based on the following equation:
[GRAPHIC] [TIFF OMITTED] TP13AP99.118
[GRAPHIC] [TIFF OMITTED] TP13AP99.332
3.5.4.3.2.3 Create a Property Net Income Multiplier
[a] The rent index series just created is combined with the
vacancy rate series (Vt) provided by the Property
Valuation component of the stress test to create a formula for
updating the average, underlying, NOI in each month of the stress
period. The following formula provides a multiplication factor that
gives the ratio of current property NOI to NOI at loan origination
(for cash programs), or at acquisition (for negotiated programs):
[GRAPHIC] [TIFF OMITTED] TP13AP99.119
[GRAPHIC] [TIFF OMITTED] TP13AP99.333
[[Page 18264]]
[b] There are two constants in the above equation. The first,
2.15, is the percentage decline in NOI due to a one percent increase
in the vacancy rate. The second, 0.0623, is the average vacancy rate
observed for multifamily rental properties in 1983-95. The average
vacancy rate is used to approximate the vacancy rate of each loan at
the time of origination (cash programs) or acquisition (negotiated
programs). Nt measures how changes in rental inflation
and vacancy rates together translate into percentage changes in net
operating income since loan origination.
3.5.4.3.3 Create a DCR Time Series
[a] DCR is the ratio of the property NOI to the mortgage
payment. DCR at loan origination or acquisition (DCR0) is
a loan characteristic input to the stress test. It is updated over
time using the formula for Nt, and by updating the
mortgage payment, if and when applicable. The mortgage payment
changes regularly for ARMs. The stress test also changes mortgage
payments for balloon loans that do not pay off at maturity. For such
loans, the coupon interest rate is changed to the prevailing market
rate at the time of balloon maturity. DCR0 for loans
purchased under original cash programs (when PR=1) of the
Enterprises are adjusted to make them consistent with current cash
programs (current measurement practices) by multiplying them by
0.8655.\12\ This adjusts for differences in appraisal practices
between original and current cash programs.
---------------------------------------------------------------------------
\12\ For Fannie Mae, these are cash loans purchased prior to
1988. For Freddie Mac, these are cash loans purchased prior to 1992.
---------------------------------------------------------------------------
[b] In addition, because UPB is decremented over time, according
to the coupon rate and amortization term for each loan group,
updates to UPB are required to update payments on ARM and balloon
loans at maturity. Updates to UPB are also used to create current
LTVs. Procedures for creating a time series of LTV ratios follows
this discussion involving DCR construction. In the following
procedures, both UPB and mortgage payments (PMT) are factors based
on an original loan balance of one dollar and do not represent
actual dollar amounts.
3.5.4.3.3.1 Create the Original Payment Factor for All Loans
The original payment factor is based on original loan terms:
[GRAPHIC] [TIFF OMITTED] TP13AP99.120
[GRAPHIC] [TIFF OMITTED] TP13AP99.334
3.5.4.3.3.2 Create Time Series of UPB Values for Fixed-rate, Fully
Amortizing Loans
For all fixed-rate, fully amortizing loans, create the UPB time
series in the stress test period according to the following
equation:
[GRAPHIC] [TIFF OMITTED] TP13AP99.121
3.5.4.3.3.3 Update Mortgage Payment Factors and UPB for ARMs and
Balloon ARMs
[a] Updating UPBt and PMTt for ARMs
requires first creating the coupon interest rate series
(rc,t) for each ARM loan group. This series will capture
the effect of period and lifetime caps on the path of coupon rates.
1. The current coupon rate at the start of the stress period,
rc,s, is used for the mortgage coupon rates in the first
12 months of the stress period rc,t:
rc,t = rc,s, for t = {1,...,12}
2. In every twelfth month, compare:
rc,t >< (rb,t + 0.02375), for t = {12, 24,
36,...108}
[GRAPHIC] [TIFF OMITTED] TP13AP99.335
3. When, upon evaluation in step 2, rc,t <
(rb,t + 0.02375), set:
rc,t+1...t+12 = min{(rb,t + 0.02375),
(rc,t + 0.01), (rc,0 + 0.05)}
[[Page 18265]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.336
4. When, upon evaluation in step 2, rc,t >
(rb,t + 0.02375), set:
rc,t+1...t+12 = max{(rb,t + 0.02375),
(rc,t-0.01), (rc,0-0.05)}
5. When, upon evaluation in step 2, rc,t =
(rb,t + 0.02375), set:
rc,t+1...t+12 = rc,t
[b] The UPB percent at the start of the stress test is
calculated using an original loan balance of one dollar, remaining
term, and an average of the origination and starting coupons. The
resulting UPB percent is used to calculate the payment factor in
month one of the stress period:
[GRAPHIC] [TIFF OMITTED] TP13AP99.122
[GRAPHIC] [TIFF OMITTED] TP13AP99.123
[GRAPHIC] [TIFF OMITTED] TP13AP99.124
[GRAPHIC] [TIFF OMITTED] TP13AP99.125
[GRAPHIC] [TIFF OMITTED] TP13AP99.337
[c] The time series of mortgage coupon rates (rc,t)
from steps 1-5 is used to generate time series of payment factors
and UPB percent factors for the remaining months of the stress
period. These two series are developed simultaneously. In each
month, each series is updated based on what happened in the other
series in the previous month:
[GRAPHIC] [TIFF OMITTED] TP13AP99.126
[[Page 18266]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.127
[GRAPHIC] [TIFF OMITTED] TP13AP99.338
3.5.4.3.3.4 Create Payment and UPB Factors for Fixed-Rate Balloons
Payment factors for balloon loans with fixed interest rates are
held constant at PMT0 until the loans reach maturity. At
maturity, the payment factor is updated to reflect current market
interest rates, the remaining loan balance, and a new amortization
term.\13\ Payment factors and UPB for balloon ARMs are constructed
using the procedures just described for ARM loans, rather than the
instructions for fixed-rate balloon loans.
---------------------------------------------------------------------------
\13\ The remaining life of the loan is reset to equal the
amortization term of the loan at origination.
---------------------------------------------------------------------------
1. Set balloon term in months, Tm, according to
product types listed in Table 3-18.
[GRAPHIC] [TIFF OMITTED] TP13AP99.270
2. Create UPBt and PMTt throughout the
stress period, according to when the balloon matures in the stress
period. Loan group UPBs are reduced according to default and
prepayment (balloon payoffs) rates (see section 3.5.4.3.6,
Calculation of Default and Prepayment Rates, of this Appendix) in
the balloon year, and for up to five years beyond the month of
balloon maturity. Loan groups with balloon maturity prior to the
start of the stress test are terminated after three years in the
stress period (thirty-seventh month). Loan groups that mature during
the stress test are terminated five years after maturity.
a. If balloon term, Tm, is less than or equal to
mortgage age at the start of the stress test, As, i.e.,
the loan has passed its balloon date or is just maturing when the
stress test begins, then UPBt and PMTt are
updated as follows:
[GRAPHIC] [TIFF OMITTED] TP13AP99.128
[GRAPHIC] [TIFF OMITTED] TP13AP99.129
[[Page 18267]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.130
[GRAPHIC] [TIFF OMITTED] TP13AP99.339
b. If balloon term, Tm, is greater than mortgage age
at start of stress test, As, then update UPBt
and PMTt as follows.
[GRAPHIC] [TIFF OMITTED] TP13AP99.131
[GRAPHIC] [TIFF OMITTED] TP13AP99.132
[GRAPHIC] [TIFF OMITTED] TP13AP99.133
[GRAPHIC] [TIFF OMITTED] TP13AP99.134
[GRAPHIC] [TIFF OMITTED] TP13AP99.340
3.5.4.3.3.5 Formula for Constructing the DCR Time Series
The formulas for updating DCR over time in the stress period are
described below.
1. For loans originated under current cash programs (where
PR=0), and for all negotiated programs:
[GRAPHIC] [TIFF OMITTED] TP13AP99.135
2. For loans originated under original cash programs, where
PR=1:
[GRAPHIC] [TIFF OMITTED] TP13AP99.136
[[Page 18268]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.341
3.5.4.3.4 Create an LTV Time Series
LTV is the ratio of the unpaid principal loan balance (UPB) to
the value of the property. The UPB is updated over time as described
above. The value of the property is adjusted based on the property
net operating income multiplier (Nt) and a capitalization
rate multiplier (described below). As with DCR, LTV must be adjusted
for loans purchased under original Enterprise cash programs, to make
them consistent with current cash programs.
3.5.4.3.4.1 Updating the Capitalization Rate Multiplier
[a] The capitalization rate multiplier is the reciprocal of the
capitalization rate and reflects what investors are willing to pay
for an annual cash flow stream on a property, given the property and
market conditions, as well as the opportunity cost of capital. LTV
is updated in the stress test according to changes in the multiplier
that result from changes in the opportunity cost of capital, as
reflected through changes in market interest rates.
[b] The capitalization rate multiplier is updated in two steps,
based on changes in the ten-year CMT yield (a proxy for changes in
the opportunity cost of capital).
1. Compute the average monthly ten-year CMT yield for the loan
origination-year:
[GRAPHIC] [TIFF OMITTED] TP13AP99.137
[GRAPHIC] [TIFF OMITTED] TP13AP99.342
2. Compute the time series of ratios of capitalization rate
multipliers based on the relative spread between the origination-
year ten-year CMT and each of the monthly values of the ten-year CMT
throughout the stress period:
[GRAPHIC] [TIFF OMITTED] TP13AP99.138
[GRAPHIC] [TIFF OMITTED] TP13AP99.343
3.5.4.3.4.2 Construct the LTV Time Series
[a] For loans acquired through current cash programs (where
PR=0), or through negotiated programs:
[GRAPHIC] [TIFF OMITTED] TP13AP99.139
[b] For loans acquired through original cash programs, where
PR=1:
[GRAPHIC] [TIFF OMITTED] TP13AP99.140
[[Page 18269]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.344
[c] For all loans, prevent LTVt from approaching zero by
resetting small values to 0.01:
[GRAPHIC] [TIFF OMITTED] TP13AP99.141
3.5.4.3.5 Compute Joint Probability of Negative Equity and Negative
Cash Flow
[a] The values of the joint probability of negative equity and
negative cash flow (JPt) are computed as the area under a
bivariate standard normal density function. The form for this
function is:
[GRAPHIC] [TIFF OMITTED] TP13AP99.142
[GRAPHIC] [TIFF OMITTED] TP13AP99.345
[b] In the calculations of JPt, the two standard
normal random variables (x and y) represent transformations of DCR
and LTV values for individual properties. Standard normal random
variables have normal (Gaussian) distributions, with a mean of zero
and standard deviation of one. Any normally distributed random
variable can be ``standardized'' by subtracting the mean from the
variable, and then dividing by the standard deviation. In this
application, the ``sample'' group for which the standard deviations
apply could include all multifamily properties in the geographic
location of the properties underlying the loan group being studied.
Here the normally distributed variables are the true, but unknown ln
(DCR) and ln (LTV) values for each loan, and their mean values are:
[GRAPHIC] [TIFF OMITTED] TP13AP99.143
[GRAPHIC] [TIFF OMITTED] TP13AP99.346
and
[GRAPHIC] [TIFF OMITTED] TP13AP99.144
[c] The limits of integration (a and b) represent the distance
between the logs of the at-risk boundaries for underlying
properties--DCR=1.00 and LTV=1.00 and--Dt and
Lt respectively. The joint probability variable is then
the value of the bivariate density function, evaluated at particular
values of the integration limits in each month of the stress period:
[[Page 18270]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.145
[d] The following steps describe how to calculate the values of
at and bt.
1. First, compute the standard deviation of ln(DCRt)
and ln (LTVt):
[GRAPHIC] [TIFF OMITTED] TP13AP99.146
[GRAPHIC] [TIFF OMITTED] TP13AP99.347
2. The limits of integration in each month of the stress test,
at and bt, are:
[GRAPHIC] [TIFF OMITTED] TP13AP99.147
[GRAPHIC] [TIFF OMITTED] TP13AP99.148
[GRAPHIC] [TIFF OMITTED] TP13AP99.348
These equations reduce to:
[GRAPHIC] [TIFF OMITTED] TP13AP99.140
[e] The coefficient of correlation between the logarithms of DCR
and LTV is: = -0.5975. It should be noted that standard
software packages that compute bivariate normal probabilities do
their integrations over the left tails of both (x and y)
distributions. To estimate the left tail of the lnDCR and the right
tail of the lnLTV distribution which is required to estimate
JPt, one simply reverses the signs on the lnLTV
integration limit (from b to -b) and the correlation coefficient
(from -0.5975 to 0.5975).
3.5.4.3.5.1 Balloon Maturity Risk (BJPt)
[a] The balloon year is defined as the 12 months leading up to
and including the maturity month. Because of the contractual
requirement to pay off a loan at maturity, a balloon loan with weak
financials is more likely to default in the balloon year than at any
previous time. The stress test captures this additional credit risk
for balloon loans by giving extra weight to the JPt
variable in the balloon year. This is accomplished by including a
second JPt term in the default equations, which is only
used for balloon loans, in the balloon year:
[GRAPHIC] [TIFF OMITTED] TP13AP99.150
[GRAPHIC] [TIFF OMITTED] TP13AP99.349
[b] Not all loans will pay off or default by balloon maturity.
For those that continue beyond balloon maturity, the stress test
updates PMTt after the balloon date with current market
interest rates (as described earlier) to simulate any increase (or
decrease) in payments upon refinancing the property. This change in
loan payments changes the default risk in the post-balloon period.
[[Page 18271]]
3.5.4.3.5.2 Relative Spread Variables (RSt,
RSDt, RSUt)
The incentive to prepay a mortgage because of the ability to
refinance at lower interest rates is proxied by relative interest
rate spreads. The difference here is that, for fixed-rate mortgages,
the relative spread is split into two variables: one for when market
rates are below the coupon rate (RSDt), and one for when
market rates are above the coupon rate (RSUt).
RSDt captures in-the-money prepayment options, and
RSUt captures any dampening effect on cash-out
refinancing when the prepayment option is out-of-the-money. For ARM
loans, the relative spread variable (RSt) compares the
current coupon rate to the current market rate on fixed-rate
products.
1. For each ARM loan group, compute the relative spread as:
[GRAPHIC] [TIFF OMITTED] TP13AP99.151
2. For each fixed-rate loan group (including balloons), create
the two spread variables:
[GRAPHIC] [TIFF OMITTED] TP13AP99.152
[GRAPHIC] [TIFF OMITTED] TP13AP99.153
3.5.4.3.5.3 Years-To-Go in the Yield-Maintenance Period
(YTGt)
[a] One feature common to most fixed-rate multifamily mortgages,
whether balloon or fully amortizing, is the yield maintenance period
(YMP). During a yield maintenance period, prepayment is restricted
because borrowers cannot prepay the mortgage without incurring
substantial penalties. For fixed-rate fully-amortizing mortgages,
the YMP is 120 months. For fixed-rate balloon loans, the YMP
averages two-thirds of the loan term, up to a maximum of 120 months.
ARM loans do not have yield maintenance periods. Table 3-19, of this
Appendix provides the term of the YMP for each loan product as
follows:
[GRAPHIC] [TIFF OMITTED] TP13AP99.271
[b] The YMP is used to create the explanatory variable years-to-
go (YTGt), which measures the number of years remaining
in the yield maintenance period of the mortgage. This explanatory
variable is a proxy for the size of prepayment penalties, which
decline throughout the YMP:
[[Page 18272]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.154
[c] YTGt has its maximum value in the first month of
loan life, and declines to zero by the end of the YMP. For loan
programs with lockouts, which prohibit prepayment for a stated time
period, YTGt is set to ten for the duration of the
lockout period.
[GRAPHIC] [TIFF OMITTED] TP13AP99.155
3.5.4.3.5.4 Relative Spread Variables in the Pre-balloon Period
(RSD1t, RSD2t)
For balloon loans during the post-yield-maintenance and pre-
balloon period, borrowers must decide whether to lock in a current
interest rate or take their chances regarding what the market rate
will be when the loan matures. To capture the additional incentive
of borrowers to prepay in the two years prior to the balloon date,
to take advantage of favorable interest rates when they exist, the
stress test provides extra weight to the RSDt variable in
both the year preceding the balloon year, and the year just prior to
that:
[GRAPHIC] [TIFF OMITTED] TP13AP99.156
[GRAPHIC] [TIFF OMITTED] TP13AP99.157
[GRAPHIC] [TIFF OMITTED] TP13AP99.350
3.5.4.3.5.5 Market Rate for Fixed-Rate Mortgages (rf,t)
The current market interest rate on fixed-rate single family
mortgages is used to capture the effect of expectations of ARM
borrowers with respect to future interest rate movements. This is in
addition to the relative spread variable, RSt, used in
the prepayment equation for ARM loans. While RSt measures
differences between long-term and short-term interest rates, the
long-term interest rate itself (rf,t) indicates the
absolute level of interest rates.
3.5.4.3.5.6 Probability of Qualifying for Refinancing at Balloon
Maturity (PQt)
[a] When a balloon loan matures, the borrower is contractually
required to pay off the outstanding UPB. To do this, the borrower
generally obtains a new loan. In practice, payoff rates are
dependent on the ability of the borrower and property to qualify for
a new loan. For multifamily mortgages, the LTV must generally be
less than or equal to 0.80, and the DCR must be greater than or
equal to 1.20. The need for the property financials to meet
origination underwriting criteria at the balloon date adds to
extension risk, i.e., the risk that the loan will not pay off, but
remain outstanding.
[b] The stress test captures extension risk at the balloon date
by estimating a separate payoff equation for balloon loans at or
beyond maturity. The payoff equation includes only one variable, the
probability of qualifying for refinancing (PQt). This is
constructed like the joint probability of negative equity and
negative cash flow variable (JPt), except that the limits
of integration now reflect the minimal requirements for loan
qualification rather than the boundary points for default. The
integration limits are from t to +
for lnDCRt (right tail) and from - to
bt for lnLTVt (left tail), where:
[GRAPHIC] [TIFF OMITTED] TP13AP99.158
[[Page 18273]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.159
[GRAPHIC] [TIFF OMITTED] TP13AP99.351
[c] The range of the integration limits is reversed from that
used in calculating the JPt variable, because
PQt is calculating the probability of financially strong
loans, while JPt calculates the probability of
financially weak loans. Again, in using a standard software package
to calculate PQt, set the integration limit for
t = -t and =
- because the package is set up to integrate left tails
only.
3.5.4.3.5.7 Loan-to-Value Ratio (LTVt)
The current loan-to-value ratio is used to capture the
propensity of investors to initiate cash-out refinancing to increase
borrowers' returns on equity. The time series of LTVt is
used as an explanatory variable in prepayment equations.
3.5.4.3.5.8 Summary of All Explanatory Variables
Table 3-20 outlines all of the explanatory variables that are
used to calculate default and prepayment rates.
[[Page 18274]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.272
3.5.4.3.6 Calculation of Default and Prepayment Rates
Conditional default and prepayment rates are calculated for each
multifamily loan group based on the explanatory variables described
above, and using statistical regression coefficients estimated on
historical data. The regression coefficients provide weighting
factors for each explanatory variable. The variables are each
multiplied by their associated regression-coefficient (weights), and
then added together to yield total weighting factors. Default and
prepayment total weighting factors are combined in pairs to
calculate the annual-equivalent conditional default and prepayment
rates for each corresponding loan group in each month of the stress
period. These annual-equivalent rates are then converted into
monthly rates.
3.5.4.3.6.1 Combining Explanatory Variables into Total Weighting
Factors
3.5.4.3.6.1.1 Default Weighting Factors (t)
The calculation of the total weighting factors for defaults
varies by loan program. Two total weighting factors are calculated
for loan defaults. One calculation is for mortgages purchased
through cash programs, and the other is for mortgages acquired
through negotiated programs. For each loan
[[Page 18275]]
group, the appropriate formula is used for the entire stress period.
For loan groups in cash programs:
t = -10.0191 + 1.2687 AYt-0.0790
(AYt) \2\ + 0.6203 PR-0.0829 DW + 7.8230 JPt +
2.6446 BJPt
3. For loan groups in negotiated programs:
t = -9.6418 + 1.0596 AYt--0.0633
(AYt) \2\ + 0.2627 RF + 0.6751 RA + 12.1660
JPt + 2.6446 BJPt
3.5.4.3.6.1.2 Prepayment Weighting Factors (t )
Prepayment total weighting factors are calculated using
equations that differ both by product type and life-cycle stage. For
any one loan group, one, two, or three different equations may be
used during the stress period. Figure 3-4 illustrates how the
prepayment weighting factor equations are used over the life of any
particular loan group. Each block represents one of the five
different equations for computing the prepayment total weighting
factors.
[GRAPHIC] [TIFF OMITTED] TP13AP99.375
1. Fixed-rate Mortgages (Fully Amortizing and Balloon Loans)
If the loan product is a ``fixed-rate'' or a non-ARM balloon, and
for t where
YMP At,
[GRAPHIC] [TIFF OMITTED] TP13AP99.160
2. Fully-amortizing loans, out of yield maintenance
If the loan product type is ``fixed-rate,'' and for t where
YMP < At:
[GRAPHIC] [TIFF OMITTED] TP13AP99.161
3. Balloon loans out of yield maintenance, but prior to
maturity.
When the mortgage product is a balloon with a fixed interest
rate, and for values of t where YMP < At and t < (m-11):
[[Page 18276]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.162
[GRAPHIC] [TIFF OMITTED] TP13AP99.352
4. Fully-amortizing ARMs, and balloon ARMs before maturity.
When the mortgage product is a fully-amortizing ARM, or a
balloon ARM where t < (m-11), then:
[GRAPHIC] [TIFF OMITTED] TP13AP99.163
[GRAPHIC] [TIFF OMITTED] TP13AP99.353
5. All balloon loans, on and after the maturity date.
When the mortgage product is a balloon (ARM or fixed-rate), then
the total weighting factors are calculated as:
[GRAPHIC] [TIFF OMITTED] TP13AP99.164
[GRAPHIC] [TIFF OMITTED] TP13AP99.354
Balloon loans do not all terminate at the balloon date. The
stress test allows them to run-off according to default and
prepayment (payoff) rate calculations, in the balloon year, and for
up to five years beyond the balloon date. All balloon loans that do
not terminate within five years beyond the balloon date are
terminated in the sixty-first month. Loan groups with balloon dates
prior to the start of the stress test (m < 0) are terminated in the
thirty-seventh month of the stress period.
3.5.4.3.6.1.3 Calculating Annual Equivalent Default and Prepayment
Probabilities
[a] Once the time series of default and prepayment total
weighting factors are computed for each loan group, they are
combined in multinomial logit equations to calculate the annual-
equivalent default and prepayment probabilities. These probabilities
represent what would happen over the course of a year, were default
and prepayment probabilities for a given month (t) to continue for
an entire year.
[b] The annual-equivalent default probability, ADt,
in each month, t, is computed as:
[GRAPHIC] [TIFF OMITTED] TP13AP99.165
and the annual-equivalent prepayment probability, APt, in
each month (t) is computed as:
[GRAPHIC] [TIFF OMITTED] TP13AP99.166
3.5.4.3.6.1.4 Terminating Balloon Loans after Maturity
At the final termination point, annual-equivalent probabilities
of default and payoff are calculated as functions of two
explanatory-variable probabilities: the joint probability of
negative equity and negative cash flow (JPt), and the
probability of qualifying for a refinancing (PQt):
[[Page 18277]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.167
[GRAPHIC] [TIFF OMITTED] TP13AP99.355
3.5.4.3.7 Calculating Monthly Default and Prepayment Rates The
monthly conditional default and prepayment rates are derived from
the annual-equivalent probabilities for each month using geometric
means. For default rates:
[GRAPHIC] [TIFF OMITTED] TP13AP99.168
and for prepayment rates:
[GRAPHIC] [TIFF OMITTED] TP13AP99.169
3.5.4.4 Output
The 120 monthly default and 120 monthly prepayment rates are
generated for each loan group and are used by the Cash Flow
component of the stress test to compute monthly dollar amounts of
loans that prepay and default (see section 3.9, Cash Flows, of this
Appendix).
3.5.5 Multifamily Loss Severity
3.5.5.1 Overview
Loss severity is the net cost to an Enterprise of a loan
default. The loss severity rate is expressed as a percentage of the
UPB at time of default. The stress test calculates loss severity
rates for each multifamily loan group for each month of the stress
period. Loss severity rates are discounted to calculate an effective
loss rate in the month of default, adjusting various cost and
revenue components of loss severity that occur following the default
date. The effective loss severity rate is multiplied by the
corresponding mortgage default rate to calculate the loan group
loss-rate. The loss-rate is multiplied by the UPB in each month to
compute the dollar amount of credit losses for each loan group.
3.5.5.2 Inputs
[a] The following loan group characteristics are used:
Program type
Portfolio
Net yield (the variable ``ry'' in equations
below) \14\
---------------------------------------------------------------------------
\14\ Net yield at the start of the stress test is used
throughout the stress period for all loan groups, including ARMs.
---------------------------------------------------------------------------
Passthrough rate (the variable ``rp'' in
equations below) \15\
---------------------------------------------------------------------------
\15\ Passthrough rate at the start of the stress test is used
throughout the stress period for all loan groups, including ARMs.
---------------------------------------------------------------------------
[b] The six-month Federal agency cost of funds (variable
``rd,t'') interest rate series is used for discounting
default-related cash flows in loss severity calculations. This
series is an output from section 3.3, Interest Rates, of this
Appendix.
3.5.5.3 Procedures
The loss severity rates are calculated by program type and
portfolio. Cash flows are discounted semi-annually. The impact of
credit enhancements on cash programs with recourse and FHA-insured
loan programs is calculated below. Credit enhancements for other
multifamily program types are applied in section 3.9, Cash Flows, of
this Appendix.
3.5.5.3.1 Retained Portfolio: Cash Programs Without Recourse
[a] The basic loss severity equation is for loan groups
consisting of retained loans purchased under cash programs without
recourse. For these loan groups, loss severity rates are calculated
as the UPB at the time of default (represented by the ``1'' in the
following equation), plus the present value of foreclosure costs and
property operating expenses, minus the net proceeds from sale of the
property:
[GRAPHIC] [TIFF OMITTED] TP13AP99.170
[[Page 18278]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.356
[b] Each NPVt value represents the loss severity rate
for loans defaulting in month t of the stress period. The timing of
events (e.g., time from default to foreclosure, etc.) used in the
equation shown above is also used in the loss severity rate
equations for all other program types and portfolios. The net
operating loss on foreclosed properties for the 13 months that the
property would be real estate owned (REO) is expensed in the seventh
month of the 13-month holding period.
3.5.5.3.2 Sold Portfolio: Programs Without Recourse or Repurchase
There is a slight change in the basic loss severity equation
shown above for sold loans purchased under cash programs without
recourse, and for negotiated programs without repurchase. Four
months of interest are passed through to investors before the loans
are bought out of security pools for default resolution. The
passthrough interest expense in the second term of the loss severity
equation, below, is discounted for two months. This represents a
midpoint of the period of interest expenditures. In addition, the
UPB at time of default is a direct cash outlay, occurring four
months after default. Therefore, the UPB at time of default is
discounted because the stress test accounts for this payment in the
month of default. Therefore, the following modified equation is
applied to sold loans purchased under cash programs without
recourse, and negotiated programs without repurchase:
[GRAPHIC] [TIFF OMITTED] TP13AP99.171
[GRAPHIC] [TIFF OMITTED] TP13AP99.357
3.5.5.3.3 Retained Portfolio: Cash Programs With Recourse
When loans are purchased under cash programs with recourse, the
seller/servicer shares any losses with the Enterprise. The stress
test computes the amount of recourse and reduces the gross severity
rate as described below.
1. Compute two additional revenue elements: interest income paid
by the seller/servicer to the Enterprise (II) and (additional)
proceeds from the seller/servicer (SP) recourse.
a. Calculate mortgage interest income, II, paid by the seller/
servicer during the time between default and foreclosure:
[GRAPHIC] [TIFF OMITTED] TP13AP99.172
[GRAPHIC] [TIFF OMITTED] TP13AP99.358
b. Calculate proceeds from the seller/servicer recourse (SP).
Calculate the seller/servicer share of loss, S, as a
fraction of the UPB:
[GRAPHIC] [TIFF OMITTED] TP13AP99.173
[[Page 18279]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.359
Reduce seller/servicer loss share (S) by the interest
income it has already paid to the Enterprise (II). Thus, the final
seller/servicer payment will be:
[GRAPHIC] [TIFF OMITTED] TP13AP99.174
2. Calculate net present value loss severity rates for defaults
in each month (t) by summing the discounted values of all cost and
revenue elements:
[GRAPHIC] [TIFF OMITTED] TP13AP99.175
In this equation, interest income (II) is discounted from the mid-
point of the time between default and foreclosure, to reflect that
interest payments are made monthly by the seller/servicer throughout
this period. The seller/servicer's payment, or share of loss, is
discounted from the foreclosure date. This is also a midpoint date,
because seller/servicers pay the Enterprise some recourse amounts
prior to foreclosure, and the rest of the recourse amount
approximately two months after foreclosure.
3.5.5.3.4 Sold Portfolio: Cash Programs with Recourse
The steps for computing loss severity rates for cash programs
with recourse for sold loans purchased follow the steps outlined for
similar programs for retained loans. The differences are that the
UPB at time of default is discounted, and there is an added expense
element, the interest passthrough expense (IE) of payments made by
the Enterprise to security holders. The UPB at time of default is
discounted because this amount is disbursed to security holders four
months after the time of default. The interest expense is computed
for four months and discounted for two months.
1. Calculate four months of passthrough interest expense:
[GRAPHIC] [TIFF OMITTED] TP13AP99.176
[GRAPHIC] [TIFF OMITTED] TP13AP99.360
2. Calculate the loss severity rate for defaults in each month,
t, using IE and other components as described above:
[GRAPHIC] [TIFF OMITTED] TP13AP99.177
3.5.5.3.5 Sold Portfolio: Negotiated Programs with Repurchase
In the case of default on negotiated programs with seller/
servicer repurchase provisions, the Enterprises' losses represent a
combination of foreclosures and alternative resolutions. These
alternatives are loan restructuring, note sales, pre-foreclosure
property sales, or acceptance of deeds in-lieu-of foreclosure.
Seller/servicers are responsible for all resolution processes,
including all post-foreclosure property
[[Page 18280]]
management and disposition. The Enterprise pays the seller/servicer
claim, C, that results from the default-resolution expenses. There
is typically a recourse account established for this purpose. Thus:
[GRAPHIC] [TIFF OMITTED] TP13AP99.178
[GRAPHIC] [TIFF OMITTED] TP13AP99.361
In this equation, the discount time period for the single cost
component is the expected time to foreclosure rather than time to
final property sale, to reflect a balance of default-resolution
types and associated time intervals before claims are filed with the
Enterprise.
3.5.5.3.6 FHA-insured Programs
Loss severities on FHA-insured mortgages are set to three
percent to reflect the costs of assigning defaulted loans to HUD.
3.5.5.4 Output
The 120 monthly loss severity rates for each loan group are used
by the Cash Flow component of the stress test to calculate monthly
amounts of credit losses, net of recourse offsets (see section 3.9,
Cash Flows, of this Appendix).
3.6 Other Credit Factors
3.6.1 Overview
The Other Credit Factors component of the stress test accounts
for sources of credit risk other than the risk of default by
mortgage borrowers. These sources of credit risk include the risk of
default by credit enhancement and derivative counterparties, as well
as the risk of default of corporate securities, municipal
securities, and rated mortgage-related securities. The stress test
classifies these sources of credit risk into four ratings categories
(``AAA'', ``AA'', ``A'' and ``BBB'') based on public ratings
information, and establishes credit loss factors appropriate to each
of these categories that are applied during the stress period.
3.6.2 Input
The stress test uses credit ratings issued by Standard & Poor's,
Moody's, Duff & Phelps and Fitch as the basis to assign
counterparties (except seller/servicers) and securities into one of
the four rating categories. The stress test only uses Standard &
Poor's and Moody's ratings for seller/servicers.
3.6.3 Procedures
3.6.3.1 Identifying Other Credit Factors
The stress test first identifies all non-mortgage borrower
sources of credit risk and associated financial instruments, and
groups them into two major categories-counterparties and securities.
Counterparties are mortgage insurers, pool insurers, seller/
servicers, and counterparties for derivative contracts. Securities
include mortgage-related securities, such as mortgage revenue bonds
(MRBs) and private label REMICs, and non-mortgage investments, such
as corporate and municipal bonds and asset-backed securities (ABSs).
3.6.3.2 Classifying Rating Categories in the Stress Test
[a] Public ratings of a counterparty or security determine the
extent of associated credit losses during the stress period. Based
on these ratings, the stress test classifies counterparties and
rated securities into one of the four rating categories:
AAA--all securities/counterparties rated between AAA/
Aaa and AAA-/Aaa3
AA--all securities/counterparties rated AA+/Aa1 and AA-
/Aa3
A--all securities/counterparties rated A+/A1 and A-/A3
BBB--all securities/counterparties rated BBB+/Baa1 and
below (Unrated corporate securities and counterparties are included
in the BBB category.)
[b] For loans with more than one layer of mortgage credit
enhancement coverage, only the ratings of the counterparty providing
the primary layer of coverage are used. If the security or the
primary coverage provider has different ratings from different
rating agencies, i.e., a ``split rating,'' then the lower rating is
used.
3.6.3.3 Accounting for Other Credit Factors
[a] The stress test specifies the final haircuts (i.e., the full
amount of discount for other sources of credit risk in the stress
period) by rating categories as shown in Table 3-21. The stress test
further specifies that haircuts increase by equal amounts in each
month until the final haircut is reached during the 120th month of
the stress period.
[GRAPHIC] [TIFF OMITTED] TP13AP99.273
[b] Haircuts for each credit rating category in each month of
the stress period can be obtained from the following formula:
[GRAPHIC] [TIFF OMITTED] TP13AP99.179
[[Page 18281]]
[GRAPHIC] [TIFF OMITTED] TP13AP99.362
[c] The haircut is applied to the cash flows of a rated security
or payments due from a counterparty according to the following
formula:
[GRAPHIC] [TIFF OMITTED] TP13AP99.180
[GRAPHIC] [TIFF OMITTED] TP13AP99.363
3.6.4 Output
The outputs of the Other Credit Factors component are the stress
period final haircuts by rating category and by counterparty and
security category. These haircuts are inputs to section 3.7,
Mortgage Credit Enhancements; section 3.5.3, Single Family Loss
Severity; section 3.5.5, Multifamily Loss Severity; and section 3.9,
Cash Flows, of this Appendix.
3.7 Mortgage Credit Enhancements
3.7.1 Overview
For each loan group and each month of the stress period, the
stress test calculates reductions to mortgage credit losses that
reflect the effects of credit enhancements. This component
calculates the values of eight loan group characteristics relating
to credit enhancements, which are part of the Enterprises' starting
position loan group characteristics, as described in Table 3-2 of
this Appendix. These characteristics, combined with counterparty
``haircuts,'' are used in section 3.5.3, Single Family Loss Severity
and section 3.5.5, Multifamily Loss Severity, of this Appendix to
calculate loss severity rates, and in section 3.9, Cash Flows, of
this Appendix to calculate dollar reductions to credit losses.
3.7.2 Inputs
This component uses the inputs listed in section 3.7.2.1,
3.7.2.2, and 3.7.2.3 of this Appendix.
3.7.2.1 Enterprise Data on Mortgage Credit Enhancements
[a] Loan-level information on mortgage credit enhancements: