Export-Import Bank: OMB's Method for Estimating Bank's Loss Rates
Involves Challenges and Lacks Transparency (30-SEP-04,		 
GAO-04-531).							 
                                                                 
The Export-Import Bank (Ex-Im Bank) facilitates U.S. exports by  
extending credit to foreign governments and corporations, mostly 
in developing countries. The Federal Credit Reform Act requires  
Ex-Im Bank to estimate its net future losses, called "subsidy	 
costs," for budget purposes. Beginning with fiscal year 2003, the
Office of Management and Budget (OMB) significantly changed its  
methodology for estimating a key subsidy cost component: the	 
expected loss rates across a range of risk ratings of		 
U.S.-provided international credits. In response to a		 
congressional mandate, GAO agreed to (1) describe OMB's current  
and former methodologies and the rationale for the recent	 
revisions, (2) determine the current methodology's impact on	 
Ex-Im Bank, and (3) assess the methodology and how it was	 
developed.							 
-------------------------Indexing Terms------------------------- 
REPORTNUM:   GAO-04-531 					        
    ACCNO:   A12835						        
  TITLE:     Export-Import Bank: OMB's Method for Estimating Bank's   
Loss Rates Involves Challenges and Lacks Transparency		 
     DATE:   09/30/2004 
  SUBJECT:   Bank loans 					 
	     Cost analysis					 
	     Exporting						 
	     Fees						 
	     Foreign loans					 
	     Future budget projections				 
	     Loan defaults					 
	     Losses						 
	     Risk management					 
	     Statistical methods				 
	     Subsidies						 
	     Transparency					 
	     OMB Interagency Country Risk Assessment		 
	     System						 
                                                                 

******************************************************************
** This file contains an ASCII representation of the text of a  **
** GAO Product.                                                 **
**                                                              **
** No attempt has been made to display graphic images, although **
** figure captions are reproduced.  Tables are included, but    **
** may not resemble those in the printed version.               **
**                                                              **
** Please see the PDF (Portable Document Format) file, when     **
** available, for a complete electronic file of the printed     **
** document's contents.                                         **
**                                                              **
******************************************************************
GAO-04-531

                United States Government Accountability Office 

                     GAO Report to Congressional Committees

September 2004 

EXPORT-IMPORT BANK

  OMB's Method for Estimating Bank's Loss Rates Involves Challenges and Lacks
                                  Transparency

                                       a

GAO-04-531

Highlights of GAO-04-531, a report to congressional committees

The Export-Import Bank (Ex-Im Bank) facilitates U.S. exports by extending
credit to foreign governments and corporations, mostly in developing
countries. The Federal Credit Reform Act requires Ex-Im Bank to estimate
its net future losses, called "subsidy costs," for budget purposes.
Beginning with fiscal year 2003, the Office of Management and Budget (OMB)
significantly changed its methodology for estimating a key subsidy cost
component: the expected loss rates across a range of risk ratings of
U.S.-provided international credits. In response to a congressional
mandate, GAO agreed to (1) describe OMB's current and former methodologies
and the rationale for the recent revisions, (2) determine the current
methodology's impact on Ex-Im Bank, and (3) assess the methodology and how
it was developed.

GAO recommends that the Director of OMB provide affected U.S. agencies and
Congress with technical descriptions of its current expected loss
methodology and update this information when there are changes. GAO also
recommends that the Director arrange for independent review of the
methodology and ask U.S. international credit agencies for their most
complete, reliable data on default and repayment histories, so that the
validity of the data on which the methodology is based can be assessed
over time.

www.gao.gov/cgi-bin/getrpt?GAO-04-531.

To view the full product, including the scope and methodology, click on
the link above. For more information, contact Celia Thomas,
(202-512-8987), [email protected].

September 2004

EXPORT-IMPORT BANK

OMB's Method for Estimating Bank's Loss Rates Involves Challenges and Lacks
Transparency

OMB changed its method for determining expected loss rates for U.S.
international credits, with one basis being that emerging finance
literature indicated the former approach might overstate losses to the
government. While it formerly used only interest rate differences across
bonds to derive expected loss rates, it now uses corporate bond default
data, adjusted for trends in interest rates, to predict defaults and makes
assumptions regarding recoveries to estimate expected loss rates. As the
figure shows, expected loss rates fell under the new approach: they were
higher across risk rating categories in fiscal year 2002 (the last year
that the former method was used) than in fiscal year 2005. This drop has
contributed to lower Ex-Im Bank projections of subsidy costs and budget
needs.

OMB's current method for estimating expected loss rates involves
challenges and lacks transparency. Estimating such losses on developing
country financing is inherently difficult, and OMB's shift to using
corporate default data has some basis, given the practices of some other
financial institutions and limitations in other data sources. However, the
corporate default data's coverage of developing countries has historically
been limited, and their predictive value for Ex-Im Bank losses is not yet
established. OMB's method generally predicts lower defaults than the
corporate default data it used, whereas more recent corporate data show
higher default rates. At the same time, OMB has assumed increasingly lower
recovery rates, which serve to somewhat offset the lower default
expectations, but the basis for the recovery rates and the changes over
time has not been transparent. In addition, despite the method's
complexity, OMB developed it independently and provided affected agencies
with limited information about its basis or structure.

OMB Expected Loss Rates for U.S. Government International Credits by Risk
Rating Category (Present Value Basis), Fiscal Years 2002 and 2005

Percentage expected loss

40

30

20

10

0

Source: GAO analysis of OMB expected loss rates.

Contents

  Letter

Results in Brief
Background
OMB Developed New Method That Lowered Expected Loss

Rates Current OMB Methodology Has Lowered Ex-Im Bank's Projected Subsidy
Costs and Budgetary Needs Loss Estimation Involves Challenges and OMB
Methodology Is Not

Transparent Conclusions Recommendations for Executive Action Agency
Comments and Our Evaluation

1 3 6

11

22

30 39 40 40

Appendixes

Appendix I: 

Appendix II: 

Appendix III: 

                                       Appendix IV: Appendix V: Appendix VI: 

Objectives, Scope, and Methodology 42

Loss Estimation Practices of Foreign Export Credit
Agencies 49
ECAs in Canada and the United Kingdom Have Methodologies to

Estimate Future Defaults and Losses in Determining Reserve

Levels 49 ECAs in France and Germany Use Fees to Offset Loss 56 OECD
Participating Countries' Risk Assessment and Fee

Arrangement 60

Loan Loss Allowance Guidance and Select Commercial Bank Practices 61 Loan
Loss Allowance Is an Important Factor in an Institution's

Financial Condition 61 Accounting and Regulatory Guidance Are Not
Prescriptive; the Loan Loss Allowance Requires Significant Judgment 62
Regulatory Agencies and Accounting Organizations Have Been Reviewing Loan
Loss Allowance Guidance 68 Reviewed Banks Follow Basic Concepts in
Accounting and Regulatory Guidance but Vary in Allowance Methodologies 70

Interagency Country Risk Assessment System 76

Credit Reform Budgeting 79

Technical Description of OMB Model for Estimating
Expected Loss of U.S. International Credit Activities 82

                                    Contents

        Appendix VII:  Comparison of OMB Default Probabilities for Fiscal Years  
                        2004 and 2005 with Corporate Default Rates Used in OMB   
                                                Model                             90 
       Appendix VIII:    Trends in Interagency Country Risk Assessment System    
                                         Expected Loss Rates                      99 
         Appendix IX:     Comments from the Office of Management and Budget      101 
          Appendix X:           GAO Contacts and Staff Acknowledgments           103 
                                             GAO Contacts                        103 
                                         Staff Acknowledgments                   103 
Tables                  Table 1: Summary of Accounting and Regulatory Guidance   
                            Followed by Banks in Their Loan Loss Allowance       
                                              Calculation                         64 
                            Table 2: Regulatory Agencies' Risk Rating Scale       67 

Figures 	Figure 1: Figure 2:

Figure 3:

Figure 4:

Figure 5:

Figure 6:

Figure 7:

Figure 8:

Figure 9:

Components of the ICRAS Process
Trends in Interest Rate Spreads for Argentine, Russian,
and Mexican Government Bonds as Compared to U.S.
Treasury Bonds, 1999-2003
Comparison of OMB Default Probabilities for Fiscal Years
2004 and 2005 and Moody's Corporate Default Rates Used
in OMB Model for Selected Rating Categories
Comparison of OMB's Expected Loss Rates, in Present
Value Terms, for ICRAS Risk Categories 1-8, Fiscal Years
2002-2005
Ex-Im Bank Obligation of Budget Authority for New
Subsidy Costs, Fiscal Years 1992-2003
Comparison of Ex-Im Bank Exposure Fees and Expected
Loss Rates by ICRAS Category, Fiscal Years 2002 and
2005
Example of a Commercial Bank's Loan Loss Allowance
Process for Corporate Loans
Program and Finance Account Budgeting for Ex-Im Bank
under Credit Reform
Illustration of How Spread Changes Can Affect the Final
Expected Default Estimates

                                       9

                                       13

                                       19

                                     21 24

26 71 80 88

91

Figure 10: Comparison of OMB Default Probabilities for Fiscal Years 2004
and 2005 for ICRAS Category 1 with Moody's Corporate Default Rates Used in
OMB Model

Contents

Figure 11: Comparison of OMB Default Probabilities for Fiscal Years

2004 and 2005 for ICRAS Category 2 with Moody's

Corporate Default Rates Used in OMB Model 92 Figure 12: Comparison of OMB
Default Probabilities for Fiscal Years

2004 and 2005 for ICRAS Category 3 with Moody's

Corporate Default Rates Used in OMB Model 93 Figure 13: Comparison of OMB
Default Probabilities for Fiscal Years

2004 and 2005 for ICRAS Category 4 with Moody's

Corporate Default Rates Used in OMB Model 94 Figure 14: Comparison of OMB
Default Probabilities for Fiscal Years

2004 and 2005 for ICRAS Category 5 with Moody's

Corporate Default Rates Used in OMB Model 95 Figure 15: Comparison of OMB
Default Probabilities for Fiscal Years

2004 and 2005 for ICRAS Category 6 with Moody's

Corporate Default Rates Used in OMB Model 96 Figure 16: Comparison of OMB
Default Probabilities for Fiscal Years

2004 and 2005 for ICRAS Category 7 with Moody's

Corporate Default Rates Used in OMB Model 97 Figure 17: Comparison of OMB
Default Probabilities for Fiscal Years

2004 and 2005 for ICRAS Category 8 with Moody's

Corporate Default Rates Used in OMB Model 98 Figure 18: Trends in ICRAS
Expected Loss Rates for 8-Year Maturity

Credits, in Present Value Terms, Fiscal Years

1997-2005 100

Contents

Abbreviations 

EL expected loss
ECA export credit agency
Ex-Im Bank Export-Import Bank
FDIC Federal Deposit Insurance Corporation
FRB Federal Reserve Board
FASB Financial Accounting Standards Board
GAAP generally accepted accounting principles
ICRAS Interagency Country Risk Assessment System
LGD loss given default
OCC Office of the Comptroller of the Currency
OMB Office of Management and Budget
OECD Organization for Economic Cooperation and Development
PD probability of default
SEC Securities and Exchange Commission
SFAS Statement of Financial Accounting Standards
U.K. United Kingdom

This is a work of the U.S. government and is not subject to copyright
protection in the United States. It may be reproduced and distributed in
its entirety without further permission from GAO. However, because this
work may contain copyrighted images or other material, permission from the
copyright holder may be necessary if you wish to reproduce this material
separately.

A

United States Government Accountability Office 

Washington, D.C. 20548 

September 30, 2004

The Honorable Richard C. Shelby

Chairman

The Honorable Paul S. Sarbanes

Ranking Minority Member

Committee on Banking, Housing and Urban Affairs United States Senate

The Honorable Michael G. Oxley Chairman The Honorable Barney Frank Ranking
Minority Member Committee on Financial Services House of Representatives

As the official U.S. export credit agency (ECA), charged with providing
financing to facilitate U.S. exports, the Export-Import Bank (Ex-Im Bank)
issues loans, guarantees, and insurance products to foreign governments
and corporations, primarily in developing countries. As of September 30,
2003, Ex-Im Bank had a portfolio of about $61 billion.1 Like any credit
institution, the bank expects that some of the credit it offers will not
be repaid, and it estimates these future losses for federal budget
purposes according to the Federal Credit Reform Act of 1990.2 The act
requires that prior to entering into loans or loan guarantees, Ex-Im Bank
must have budget authority for its "subsidy costs"-broadly speaking,
estimates of net

1This portfolio valuation includes guarantees, loans receivable,
insurance, receivables from subrogated claims, and undisbursed loans.
Claims are made to Ex-Im Bank when a loan that it has guaranteed or an
insurance policy that it has issued becomes overdue or defaults.

2Pub. L. No. 101-508.

losses on a present value basis.3 The Office of Management and Budget
(OMB) has overall responsibility for coordinating cost estimates under
credit reform and plays a unique role in determining the subsidy costs of
Ex-Im Bank and other federal agencies that offer international credit-it
provides these agencies with expected loss rates, a key component of their
subsidy costs.4 For the fiscal year 2003 budget, OMB significantly changed
its methodology for determining these rates.5 In its annual financial
statements, Ex-Im Bank also accounts for future expected losses by
establishing loss allowances in accordance with private sector accounting
standards.6

The Export-Import Bank Reauthorization Act of 2002 directed GAO to report
on the bank's "reserve practices," which include its approach for
estimating subsidy costs.7 In response, we agreed to (1) describe OMB's
current and former methodologies for estimating expected loss rates for
U.S. credit agencies' international credit and the rationale for the
recent revisions, (2) determine the impacts of the current methodology on
Ex-Im Bank, and (3) assess the current methodology and the process by
which it was developed. We also agreed to provide information on foreign
ECA and commercial bank practices for estimating expected losses.

3The Federal Credit Reform Act defines the cost of a direct loan as the
net present value (at the time of loan disbursement) of loan
disbursements, principal repayments, and interest payments, adjusted for
estimated defaults, prepayments, fees, penalties, and other recoveries. It
defines the cost of a loan guarantee as the net present value (at the time
the underlying loan is disbursed) of estimated payments by the government
(for defaults and delinquencies, interest rate subsidies, and other
payments) minus estimated payments to the government (for origination and
other fees, penalties, and recoveries). When the present value of payments
exceeds the present value of receipts-that is, when a credit program loses
money-a positive subsidy exists. When the converse is true, a negative
subsidy exists.

4The risk ratings assigned to transactions are also an important
determinant of subsidy costs. Information on how risk ratings are
determined for Ex-Im Bank is presented in the background section of this
report, although an assessment of the appropriateness of ratings was
outside the scope of this review.

5OMB made a small across-the-board downward adjustment to its expected
loss rates for the fiscal year 2002 budget, but this did not entail a
change in its basic methodology.

6Financial statement loss allowances consist of allowances for losses on
loans and claims receivable and liabilities for losses on insurance and
guarantee programs. These allowances are a measurement that reflects
probable and estimable uncollectible loan balances or potential future
liabilities, as required under private sector accounting standards. They
are not tied directly to a funding request.

7Pub. L. No. 107-189, Sec. 14 (12 U.S.C. 635 note).

To describe and assess OMB's current methodology for estimating expected
loss rates, we obtained and evaluated analytical papers and OMB data and
assumptions and discussed this information with OMB representatives. We
reviewed an OMB paper that described the current methodology in
theoretical terms and obtained more complete information by, on several
occasions, posing questions to technical staff through OMB's Office of
General Counsel. While we obtained sufficient information to generally
describe and assess key aspects of the methodology, we did not replicate
or validate it. We also did not determine the reasonableness of specific
loss rates that OMB has estimated. We note in the report where our
description of certain aspects of the methodology is incomplete, but these
areas were not material to our conclusions. We discussed the development
of OMB's loss estimation methodology with knowledgeable U.S. government
officials. We also reviewed relevant research and discussed key issues
with selected commercial banks, foreign ECAs and related government
agencies, and credit experts. To determine the impact of the current
methodology on Ex-Im Bank, we analyzed the bank's budget and financial
statement documents and discussed them with bank officials. Appendix I
provides a more detailed description of our scope and methodology;
appendixes II and III contain descriptions of foreign ECA and commercial
bank practices for estimating expected losses. We conducted our review
from November 2002 through March 2004 in accordance with generally
accepted government auditing standards.

Results in Brief 	OMB developed its current methodology for determining
expected loss rates, which lowered them, in part because of finance
literature indicating that its former approach likely overstated losses to
the government. OMB's former methodology for estimating loss rates relied
on interest rate differences-"spreads"-between bonds at different risk
levels and lowrisk bonds such as U.S. Treasury bonds. The former
methodology assumed that higher interest rates on bonds at different risk
levels signaled the extent to which they presented higher probabilities of
default and expected loss. The finance literature indicated that other
factors in addition to expected losses, such as tax and liquidity
considerations, influence interest rate differences. OMB's current
methodology uses rating agency corporate default data and interest rate
spreads in a model it developed to estimate default probabilities and
makes assumptions about recoveries after default to estimate expected loss
rates. This methodology has generally predicted default rates somewhat
lower than the underlying corporate rates it uses. Under the current
methodology, expected loss rates for 8-year maturity credits were on
average about 58 percent lower in fiscal year 2005 than in

fiscal year 2002, in risk categories in which Ex-Im Bank generally
undertakes new financing.

With lower loss rates, OMB's current methodology has contributed to Ex-Im
Bank projections of lower subsidy costs and budgetary requirements and
influenced a modification in the way the bank calculates loss allowances
for its financial statements. OMB's new loss rates contributed to the
bank's request of smaller budget authority in fiscal years 2003 through
2005 to cover its anticipated subsidy costs. In addition, in fiscal year
2003, Ex-Im Bank's obligation of budget authority for subsidy costs
dropped by almost half from fiscal year 2002, while the amount and
estimated average risk of the bank's new financing in those years was
similar. When Ex-Im Bank reestimated the subsidy cost of its outstanding
portfolio at the end of fiscal year 2002 using the new rates, these costs
dropped by $2.7 billion, a decrease attributed by Ex-Im Bank officials
primarily to OMB's lower loss rates. Further, the fees that Ex-Im Bank
charges to compensate for risk are now projected to generally provide
greater coverage of its expected losses. During this period, Ex-Im Bank
modified its approach for calculating financial statement loss allowances
to be more in line with applicable accounting standards. This involved,
among other things, diverging from its former practice of using the same
loss rates to calculate loss allowances and subsidy costs. To maintain
consistency in its loss allowance estimation and because of the changed
nature of OMB's loss rates, Ex-Im Bank generally began using higher loss
rates for its loss allowances than it did for its subsidy costs.

The OMB's current methodology for estimating expected loss rates for U.S.
agencies' international credits involves challenges and is not
transparent. Estimating such losses on developing country financing is
inherently difficult, and OMB's shift to using corporate default data has
some basis, given the practices of some other financial institutions and
limitations in other data sources. However, the corporate default data's
coverage of developing countries has historically been limited, and their
predictive value for Ex-Im Bank losses is not yet established. More recent
corporate default data than what OMB used shows higher defaults in some
risk categories. In deciding to use this data to predict default, OMB
analyzed Ex-Im Bank historical defaults over a somewhat narrow period. The
default data analyzed did not cover other U.S. international credit
agencies. OMB's recovery rate assumptions have dropped twice since the
methodology was implemented. The lower rates serve to offset lower default
projections in the overall estimation of expected loss, but the basis for
the recovery rates and the changes over time has not been transparent.
Finally, despite the

complexity and implications of the current methodology, OMB developed it
independently and provided affected agencies with limited information
about its basis or structure.

To improve the transparency of the subsidy cost estimation process and
help ensure its validity, we are recommending that the Director of the
Office of Management and Budget take five actions. First, we recommend
that the Director provide affected U.S. agencies and Congress technical
descriptions of OMB's current method of determining expected loss rates.
Second, we recommend that the Director provide similar information in the
event of significant changes to its method for calculating expected loss
rates. Third, we recommend that the Director ensure that OMB periodically
update data from nonagency sources, such as the corporate default data
used to estimate expected loss rates. Fourth, we recommend that the
Director request from Ex-Im Bank and other U.S. international credit
agencies the most complete and reliable information available on their
default and repayment histories, so that the validity of the information
on which the current methodology is based can be assessed over time.
Finally, we recommend that the Director provide for, and document,
independent methodological review of OMB's expected loss model.

Commenting on a draft of this report, OMB generally agreed to implement
these recommendations. OMB also expressed concern about the report's
statement that its method for determining loss rates was not transparent,
observing that our report generally describes the method. We believe that,
while we do present in this report a substantial amount of information on
OMB's loss methodology, obtaining that information required considerable
resources and effort, and similar information should be more readily
available on an ongoing basis to affected agencies and Congress. Ex-Im
Bank and the Comptroller of the Currency reviewed the report and made
technical comments, which we incorporated where appropriate. The
Department of the Treasury, the Federal Reserve Board, and the Federal
Deposit Insurance Corporation did not have comments on the report. The
Securities and Exchange Commission and the Department of Agriculture's
Foreign Agricultural Service reviewed parts of the report for technical
accuracy; the Securities and Exchange Commission provided technical
comments, which we incorporated where appropriate. We also obtained
technical comments from bank and foreign ECA officials on our descriptions
of their practices.

Background 	Established in 1934, Ex-Im Bank is an independent U.S.
government corporation that serves as the official ECA of the United
States.8 Its mission is to support the export of U.S. goods and services
overseas, thereby supporting U.S. export sector jobs. Ex-Im Bank's mandate
states that it should not compete with the private sector but rather
assume the credit and country risks that the private sector is unable or
unwilling to accept. Ex-Im Bank offers various financial products, such as
direct loans, loan guarantees, export credit insurance, and working
capital guarantees, to foreign buyers of U.S. goods and services and to
U.S. exporters. In the last decade, new Ex-Im Bank authorizations of
loans, guarantees, and insurance averaged nearly $12 billion per year.

Because of its mandate, a large percentage of Ex-Im Bank's business is
with developing country borrowers that are typically considered more risky
than borrowers in developed countries. Nearly 80 percent of Ex-Im Bank's
medium- and long-term exposure at the end of fiscal year 2003 was to
borrowers from low-and middle-income countries.9 According to Ex-Im Bank
officials, the types of borrowers it finances within countries have
shifted over the last decade: whereas Ex-Im Bank historically financed
foreign government (sovereign) purchases of U.S. exports, its new
financing is now primarily for purchases by private sector borrowers. This
shift is gradually being reflected in Ex-Im Bank's portfolio of
outstanding credits, which at the end of fiscal year 2003 included about
36 percent in financing to sovereign governments, about 46 percent in
financing to foreign corporations, and about 18 percent in financing to
public sector, nonsovereign borrowers.

Both sovereign and private borrowers present some risk of failing to meet
payment obligations (i.e., defaulting), potentially causing a financial
loss for Ex-Im Bank and the U.S. government.10 In 1990, to more accurately
measure the cost of federal credit programs, the government enacted credit

8Ex-Im Bank, which is subject to reauthorization every 4 to 5 years, was
last reauthorized in June 2002.

9In grouping the countries for which Ex-Im Bank reported exposure in its
2003 financial statement, we used World Bank and Organization for Economic
Cooperation and Development income classifications.

10On a direct loan, default occurs when payments due to Ex-Im Bank are not
made as scheduled. On a guaranteed loan, default occurs when payments due
to the private sector lender are not made as scheduled, causing the lender
to file a claim with Ex-Im Bank.

reform, which required agencies that provide domestic or international
credit, including Ex-Im Bank, to estimate and request appropriations for
the long-term net losses, or subsidy costs, of their credit activities.11
According to credit reform, Ex-Im Bank incurs subsidy costs when estimated
payments by the government (such as loan disbursements) exceed estimated
payments to the government (such as principal repayments, fees, interest
payments, and recovered assets), on a present value basis over the life of
the loan. For each credit activity, Ex-Im Bank assesses the potential
future losses based on the risk of the activity. It collects up-front fees
or charges borrowers higher interest rates, or both, to offset that loss
and receives subsidy appropriations to cover remaining losses.

Credit reform requires credit agencies to have budget authority to cover
subsidy costs before entering into loans or loan guarantees. Credit
agencies, in their annual appropriations requests, estimate the expected
subsidy costs of their credit programs for the coming fiscal year. Credit
reform also requires agencies to annually reestimate subsidy costs of
previous financing activity based on updated information. When reestimated
subsidy costs exceed agencies' original subsidy cost estimates, the
additional subsidy costs are not covered by new appropriations but rather
are funded from permanent, indefinite budget authority.

To estimate their subsidy costs, credit agencies estimate the future
performance of direct and guaranteed loans. Agency management is
responsible for accumulating relevant, sufficient, and reliable data on
which to base these estimates. To estimate future loan performance,
agencies generally have cash flow models, or computer-based spreadsheets,
that include assumptions about defaults, prepayments, recoveries, and the
timing of these events and are based on the nature of their own credit
program. Agencies that provide credit to domestic borrowers generally
develop these cash flow assumptions, which OMB reviews, based on their
historical experiences. For U.S. international credits, OMB provides the
expected loss rates, which are composed of default and recovery
assumptions, that agencies should use to estimate their subsidy costs.

11Federal agencies that provide credit to the domestic market include the
Departments of Agriculture, Education, Housing and Urban Development, and
Veterans Affairs and the Small Business Administration. Federal agencies
that provide international credit include the Departments of Agriculture
and Defense and the Agency for International Development, Ex-Im Bank, and
the Overseas Private Investment Corporation.

The determination of expected loss rates for federal agencies that provide
international credit has two components: the assignment of risk ratings
for particular borrowers or transactions and the determination of loss
rates for each rating category, according to the maturity of the credit.12
Both of these components, and their relationship to one another, are
important in determining overall expected losses. For Ex-Im Bank, risk
ratings are determined partly through an interagency process and partly by
Ex-Im Bank's risk management division. The appropriateness of these
ratings is a key determinant in the overall appropriateness of Ex-Im
Bank's subsidy cost estimations.13

Through the Interagency Country Risk Assessment System (ICRAS),14 which
OMB chairs, ICRAS agencies determine risk ratings that will be in effect
each fiscal year (see box 1 in fig. 1).15 There are two types of ICRAS
ratings-one for foreign government (sovereign) borrowers and one for the
private sector climates in foreign countries. Ratings range from 1 (least
risky) to 11 (most risky). Ratings for sovereign borrowers are based on
macroeconomic indicators, such as indebtedness levels; balance-ofpayments
factors; and political and social factors. In determining ratings, the
agencies take into account country risk ratings assigned by private sector
ratings agencies and by the Organization for Economic Cooperation

12These expected losses are estimates, based on available information, of
the mean, or average, level of future losses expected from particular
credit activities. Actual losses can be higher or lower than the expected
losses.

13Evaluating the risk rating process or the reasonableness of specific
ratings was beyond the scope of this engagement.

14ICRAS was established in 1991 to create uniformity among the federal
agencies involved in providing international credit. According to OMB,
these agencies had previously used separate methodologies for estimating
their subsidy costs, which often produced different default expectations
for the same debtor. The ICRAS working group is chaired by OMB and
includes representatives from the cross-border financing agencies,
including Ex-Im Bank, the Departments of Agriculture and Transportation,
the Overseas Private Investment Corporation, the Agency for International
Development, and the Defense Security Assistance Administration. Other
interested government organizations, including the Departments of
Treasury, State, and Commerce; the Federal Reserve; the Council of
Economic Advisors; and the National Security Council are also represented.

15The ICRAS ratings for some countries are reviewed yearly, while others
are reviewed less frequently. Some ratings may be revised more frequently
depending on circumstances.

and Development (OECD).16 Private sector ratings assigned through the
ICRAS process also take into account factors such as the banking system
and legal environment in a country. Ex-Im Bank generally authorizes, with
few exceptions, new business for borrowers with ICRAS ratings of 8 or
better.17 (App. IV contains more information about the ICRAS risk rating
process.)

Figure 1: Components of the ICRAS Process

Sources: GAO analysis of Ex-Im Bank and OMB documents.

For Ex-Im Bank's financing with foreign governments, the ICRAS sovereign
risk rating applies. For Ex-Im Bank's private sector lending, Ex-Im Bank
officials assign risk ratings. According to Ex-Im Bank officials, they use
private rating agency ratings for a corporation when the ratings are
available, which is the case for a minority of borrowers. For most private
sector borrowers, Ex-Im Bank officials use the private sector ICRAS rating

16These comparisons are made based on a table, or concordance, that sets
up a cross-walk between ICRAS ratings and the ratings of major private
rating agencies, such as Moody's Investors Service and Standard & Poor's,
as well as between ICRAS ratings and OECD ratings.

17Ex-Im Bank's Country Limitation Schedule identifies the countries for
which the bank's support is not available or for which limitations on
available credit length exist. See
http://ww.exim.gov/tools/country/country_limits.html.

as a baseline and adjust that rating depending on their assessment of the
borrower's creditworthiness.18

For the second component, OMB plays a key role. It determines expected
loss rates for each ICRAS risk rating and maturity, which U.S. agencies
that provide international credit use in preparing their subsidy cost
estimates (see fig. 1, box 2). OMB provides these loss rates to ICRAS
agencies each fiscal year, in time to be used in preparing budget
submissions.19 To estimate future cash flows, ICRAS agencies use OMB's
expected loss rates in their cash flow models. The loss rates are also
used to allocate subsidy costs during the fiscal year and to calculate
subsidy cost reestimates at the end of the fiscal year. OMB also provides
agencies with a credit subsidy calculator, which has been audited, that
agencies use to convert agencyestimated cash flows into present values.20

The credit reform act resulted in the establishment of a special budget
accounting system to track inflows and outflows associated with agencies'
lending activities. Expected long-term subsidy costs for financing
activities in a fiscal year appear in an agency's annual budget submission
and are subject to congressional approval. However, any increases over
time in expected subsidy costs for financing that took place in earlier
years are financed from permanent indefinite budget authority and do not
have to be appropriated in the annual appropriations process.21 In the
case of Ex-Im Bank, such changes could result, for example, from changes
in the risk assessment for certain countries or changes in loss
assumptions for a given

18For medium-term transactions of less than $10 million (which represent
less than 10 percent of Ex-Im Bank's portfolio), the bank uses a portfolio
approach to assign rating categories, assigning an overall category to a
country. According to Ex-Im Bank officials, the category assigned to these
transactions is generally one risk category higher than the private sector
ICRAS rating for the country.

19For credit programs, OMB also provides the discount rates that are used
to calculate subsidy estimates. These rates are built into OMB's credit
subsidy calculator.

20PricewaterhouseCoopers LLP audited the credit subsidy calculator in
December 1999 to ensure that the calculations it is designed to make are
done correctly. The calculator was audited because users, as well as the
accountants and auditors who prepare and audit agency financial
statements, need to have assurance that it calculates reliable subsidy
costs in compliance with applicable legislation and accounting standards.

21Permanent budget authority is available as the result of previously
enacted legislation and does not require new legislation for the current
year. Indefinite budget authority is budget authority of an unspecified
amount of money.

risk level. (App. V contains additional information about the credit
reform budget accounting system.)

In addition to estimating expected losses for budgetary purposes, Ex-Im
Bank measures the expected loss of its portfolio in its own annual audited
financial statements. As a government corporation, Ex-Im Bank is required
to follow "principles and procedures applicable to commercial corporate
transactions."22 Ex-Im Bank's financial statements are prepared according
to private sector generally accepted accounting principles (GAAP) that
require Ex-Im Bank to follow Financial Accounting Standards Board (FASB)
accounting guidance when establishing allowances for future expected
credit losses.

OMB Developed New Method That Lowered Expected Loss Rates

OMB developed its current methodology for determining expected loss rates
for ICRAS agencies, which lowered these rates, based in part on evidence
that its former approach overstated likely defaults and losses. For fiscal
years 1992-2002, OMB based its expected loss estimates on differences
between interest rates on bonds of different risk levels. In developing
its current approach, OMB cited emerging academic literature that
indicated its former approach may have overestimated likely costs to the
government. Ex-Im Bank officials also said they believed, based on their
reestimates, that their subsidy cost appropriations had been too high
relative to their loss experience since the beginning of credit reform.
OMB's current approach uses historical corporate bond default data,
adjusted for trends in interest rate spreads, to predict defaults and
applies an assumption regarding recovery rates to estimate expected loss
rates. Under the current approach, loss rates across most risk categories
dropped significantly.

OMB's Former Methodology Based Expected Loss Estimates on Differences in
Bond Interest Rates

The method that OMB used in fiscal years 1992-2002 based expected loss
rates for ICRAS agencies on interest rate spreads between publicly traded
U.S. corporate or foreign government bonds and low-risk bonds such as

22The Government Corporation Control Act of 1945 required wholly owned
government corporations, including Ex-Im Bank, to follow private sector
principles and procedures. Since 1990, the act has required such
corporations to undergo annual audits by independent public accountants.

U.S. Treasury bonds.23 Under this method, estimates of expected loss
shifted as the underlying spread data shifted. Interest rate spreads are
an indicator of expected loss, in that the size of a spread tends to widen
as the perceived risk increases. For example, when interest rates on a
foreign bond are 6 percent and U.S. Treasury bond interest rates are 5
percent, the spread between the two is 1 percentage point. The foreign
bond in this example provides a higher rate of interest than the U.S.
Treasury bond because creditors require a higher return on their capital,
at least in part because they perceive that foreign bonds carry a higher
risk of nonrepayment.

Spreads fluctuate over time depending in part on changes in market views
of borrowers' creditworthiness. Figure 2 shows interest rate spreads for
Argentine, Russian, and Mexican government bonds over U.S. Treasury bonds
from 1999-2003, illustrating how spreads can fluctuate. Spreads increased
sharply in 2001 for the Argentine bonds, as Argentina's default on those
bonds was imminent. Conversely, the spread for the Russian bonds shown
narrowed over the period as Russia's economy improved, while the spreads
for the Mexican bonds were consistently the smallest of the three
countries.

23In these years, OMB presented its loss rates, for most ICRAS categories,
in terms of risk premiums, which were estimated average differences
between the interest rates on traded bonds in a risk category and the U.S.
Treasury bond rate. The default costs, or expected losses, associated with
those risk premiums were estimated to be the difference between the
present value of the loan or loan guarantee's expected cash flows
discounted at the Treasury interest rate and the expected cash flows
discounted at the risk-adjusted interest rates.

Figure 2: Trends in Interest Rate Spreads for Argentine, Russian, and
Mexican Government Bonds as Compared to U.S. Treasury Bonds, 1999-2003

Percentage points

                                       60

                                       50

                                       40

                                       30

                                       20

                                       10

                                       0

1999 2000 2001 2002 2003

Year

Argentina

Mexico

Russia

Source: GAO analysis of bond price data from Global Insight.

OMB has used varying underlying instruments to calculate bond spreads and
expected losses for ICRAS agencies. In the beginning of credit reform, OMB
used the spreads on U.S. corporate bonds at different risk levels to
estimate risk premia (and thus expected loss).24 That is, OMB determined
the interest rate spread for U.S. corporate bonds within a risk rating
category and used those spreads to compute a risk premium for each

24Initially, the spreads were computed relative to the lowest risk, or
AAA, corporate bonds. Later, OMB computed the spreads relative to U.S.
Treasury bonds.

ICRAS category. In fiscal year 1997, OMB began using the interest rate
spreads on other instruments, including foreign government bonds.25

After interest rates on some types of international bonds rose in the late
1990s, OMB determined that basing expected loss rates only on interest
rate spreads resulted in estimates that were too high. According to OMB,
it was decided in the discussions within the executive branch and with
Congress leading to credit reform that only the expected cost to the
government was relevant for estimating default losses to the government
under credit reform. OMB decided to change its method for determining
default losses, primarily because emerging research showed that factors
other than expected losses from defaults account for a significant portion
of interest rate spreads.26 According to this literature, differences in
liquidity and tax considerations, and an aspect of credit risk that OMB
termed "portfolio risk," affect interest rates on international bonds.27

Studies cited by OMB and other related literature indicate that factors
other than expected losses from defaults account for a high proportion of
interest rate spreads-in some cases, most of the spread-especially on
higher-quality bonds. For bonds with risk ratings that correspond to the
riskier ICRAS rating categories, 5 and higher (riskier), conclusions from
the literature that OMB cited and other literature that we reviewed are
less clear. One study cited by OMB found that differences in tax
treatment, and compensation for risk beyond expected losses, explained
most of interest rate spreads; however, because of limited data, that
study did not include bonds in risk categories higher than those
corresponding to ICRAS

25We reported in 1994 that expected losses using this method were based on
small numbers of spread observations in some ICRAS categories. See GAO,
Credit Reform: U.S. Needs Better Method for Estimating Cost of Foreign
Loans and Guarantees, NSIAD/GGD-95-31 (Washington, D.C.: Dec. 19, 1994).

26Literature cited by OMB included, in part, Jerome S. Fons, "Using
Default Rates to Model the Term Structure of Credit Risk," Financial
Analysts Journal 50 (1994): 25-32; and Edwin Elton, Martin J. Gruber,
Deepak Agrawal, and Chrisotopher Mann, "Explaining the Rate Spread for
Corporate Bonds," Journal of Finance 56 (2001): 247-277.

27"Portfolio risk" is the risk associated with the variability of default
rates-the likelihood that losses from defaults will be higher in some
periods than others. This portfolio risk, although related to default
costs, is not included in OMB's calculation of expected losses because,
according to OMB, it is not considered to be a cost to the government and,
thus, is not a cost for which the U.S. government would need to budget.

category 4.28 A second study cited by OMB found that market interest rate
spreads on bonds were greater than those that would be predicted based on
corporate default data. The differences were particularly apparent for
bonds in investment-grade categories and were smaller for speculativegrade
bonds.29

For the fiscal year 2002 budget, OMB imposed an across-the-board reduction
in the expected loss estimates for ICRAS risk categories 1 through 8.30
OMB said that it did this to eliminate part of the spread between other
bonds and U.S. Treasury bonds to come closer to measuring only default
cost. The risk factors and expected loss estimates for the bottom three
categories did not change.

A further rationale for adjusting the expected loss rates, according to
Ex-Im Bank officials, was that the bank had calculated several downward
reestimates of its subsidy costs since the inception of credit reform.
They viewed this as evidence that the bank's original subsidy cost
estimates were conservative. According to Ex-Im Bank officials, several
factors influence the bank's subsidy cost reestimates, including changes
in the outstanding balance of its cohorts (the term "cohort" refers to the
financing extended in a given fiscal year, which Ex-Im Bank further
subdivides by product type); changes in cohort performance or average
riskiness; and changes in OMB's expected loss rates.31 Ex-Im Bank
calculated a net downward reestimate of about $368 million in fiscal year
1999, followed by a larger net downward reestimate of about $1.4 billion
in fiscal year 2000 and a subsequent net

28Elton et.al., "Explaining the Rate Spread for Corporate Bonds."

29Fons, "Using Default Rates to Model the Term Structure of Credit Risk."

30This amount of the reduction was based on the amount by which the
default probability implied by interest rate spreads for ICRAS A-rated
credits was greater than published default probabilities for AA and
AAA-rated corporations.

31According to an Ex-Im Bank official, the bank calculates subsidy cost
reestimates at the end of the fiscal year using an approach that is
similar but not identical to one of the two approaches specified under
credit reform. Ex-Im Bank officials said that to reestimate the subsidy
cost of each cohort, they determine the outstanding principal balance at
the end of the fiscal year, the weighted average remaining term to
maturity, and the weighted average risk rating. They apply the most
current OMB expected loss rate that corresponds to each cohort's average
risk rating to the cohort's outstanding principal balance to determine the
cohort's reestimated subsidy cost. They compare that amount with the
amount already set aside for the cohort. If reestimated subsidy costs are
less than the amounts set aside the previous fiscal year, this would
result in a downward reestimate, the amount of which is transferred from
Ex-Im Bank's financing account to the Treasury.

downward reestimate of about $300 million in fiscal year 2001.32 (In these
years, upward reestimates of some cohorts were more than offset by larger
downward reestimates of other cohorts.) There were small net downward and
upward reestimates in fiscal years 1992 through 1995, and Ex-Im Bank did
not calculate reestimates in fiscal years 1996 through 1998. Ex-Im Bank's
reestimates represent the bank's ongoing assessment of the riskiness of
its post-credit reform financing at a given point in time and are not a
final assessment of the performance of cohorts that have not reached
maturity at the time of the reestimate. An Ex-Im Bank official noted that
future claims or defaults could occur on cohorts that have not reached
maturity, possibly causing upward reestimates to certain cohorts in the
future.

OMB's Current Methodology Bases Expected Loss Rates on Corporate Default
Data, Interest Rate Spreads, and Recovery Assumptions

OMB's current methodology uses rating agency corporate default data and
interest rate spreads to estimate default probabilities and makes
assumptions about recoveries after default to estimate expected loss
rates. The methodology estimates default rates for federal international
credits using a complex model that OMB developed. These rates were
generally lower for fiscal years 2004 and 2005 than the underlying
corporate default rates that OMB used in estimating its rates. OMB
introduced its current methodology, which estimates expected loss rates
for ICRAS categories 1 through 8, for use in fiscal year 2003, and made
modifications for fiscal years 2004 and 2005.33 (App. VI contains a
technical description of the methodology.)

32Information in this report about Ex-Im Bank's reestimates is based on
information that the bank provided for each year in which it calculated
reestimates. This information differs slightly in some years from
information about the bank's reestimates that is reported in the Federal
Credit Supplement to the budget. These figures reflect only the subsidy
cost portion of the reestimates cited. Ex-Im Bank also calculates interest
costs on the subsidy costs, but these costs are not included in the
figures.

33Expected losses for ICRAS categories 9 through 11 are calculated
differently. They are based on market prices (interest rates) on debt
issues of countries in these categories. According to OMB, it averaged the
limited interest rate observations of international debt for countries in
each category. OMB obtained the data from the International Finance Review
and WesBruin Capital. OMB changed its method for fiscal year 2004 (1) to
exclude collateralized instruments; (2) for performing bonds, to adjust
for the difference between bond coupon rates and Treasury rates, excluding
issues with unknown coupon rates; and (3) to apply a discount to each
ratings category (5 percent for category 9, 10 percent for category 10,
and 25 percent for category 11) to reflect that countries with high ratios
of bilateral debt (debt owed to other countries) to private debt are more
likely to expect debt reduction.

OMB Model Bases Default Estimates on Corporate Default Data and Spreads

OMB uses rating agency corporate default data and information on interest
rate spreads to determine expected defaults through a complex model. The
model has two empirical relationships, one between ratings and defaults
and the other between interest rate spreads and defaults. The model
combines the relationships to arrive at OMB's expected default rates
across ICRAS risk categories. Historical default rates on corporate bonds
by risk rating category are the key inputs to both components of the
model.

The first component of the model bases the probability that ICRAS agency
borrowers will default on default rates for corporate bonds published in
2000 by a nationally recognized private rating agency, Moody's Investors
Service. The risk categories associated with the Moody's corporate default
probabilities are converted to ICRAS risk categories.34 OMB's model uses
two Moody's data series on U.S. corporate bond defaults, which OMB
combined into a single series. The data series used for the four
lowest-risk ICRAS categories (1-4) includes default rates on rated
corporate bonds by risk rating category during 1920-1999. The data series
used for the next four (higher risk) ICRAS categories (5-8) includes
default rates on rated corporate bonds by risk rating category during
1983-1999.35

The second component of the model uses data on interest rate spreads to
make adjustments to the same Moody's historical default data. The current
method does not use spread information as the primary indicator of default
risk, as OMB's former method did. Instead, it uses spread information as a
signal of how current market conditions might differ from those reflected
in the Moody's historical data. The model is designed to adjust historical
default rates by rating category up or down in cases where interest rate
spreads in a category are unusually high or low relative to the average
spreads for that category. The adjustment in the model gives greater
weights to more recent spreads in calculating the averages. To estimate
this relationship, OMB used interest rate data on international bonds from
Bloomberg.

34Moody's ratings are themselves determined by the likelihood of default
(default rates) and the severity of default (recovery rates), according to
a Moody's official and agency documents.

35OMB converted the default probabilities in the Moody's tables to default
probabilities for ICRAS ratings by averaging certain values within each
table and making some judgmental decisions, which, according to OMB,
generally resulted in choosing values on the higher side of the Moody's
ratings when there was not a straight match with ICRAS categories.

The default probabilities reflected in OMB's expected loss rates for
fiscal years 2004 and 2005 were generally lower than the corporate default
rates that OMB used in its model. Figure 3 illustrates OMB's fiscal year
2004 and 2005 default probabilities for 1-8 years for three ICRAS ratings
categories and the Moody's corporate default rates for corresponding risk
categories.36 The graph shows that OMB default probabilities are somewhat
lower than the corporate default rates for the ratings categories shown.
(App. VII presents similar comparisons for ICRAS categories 1-8.) Based on
information we obtained on OMB's model, this difference would be expected
to result from interest rate spreads' trending significantly downward for
some rating and maturity categories. It could also result from features of
the model specification. We could not determine the reasons for the
difference because we did not replicate OMB's model and, in response to
our questions, OMB did not identify specific reasons for the differences.
(App. VI contains more information about model specification issues.)

36We compared the Moody's and OMB's default rates for maturities of 1-8
years because the combined Moody's series that OMB used had only 8 years
of default rates for ICRAS categories 5 through 8.

Figure 3: Comparison of OMB Default Probabilities for Fiscal Years 2004
and 2005 and Moody's Corporate Default Rates Used in OMB Model for
Selected Rating Categories

ICRAS score: ICRAS score: ICRAS score:

Percentage default rate or probability

50

40

30

20

10

0 12345678

Years to maturity

Methodology Combines Default Rates and Recovery Rates to Determine
Expected Losses

Percentage default rate or probability

50

40

30

20

10

0 12345678

Years to maturity

Corporate default rates used in OMB model

FY 2005 OMB default probability

FY 2004 OMB default probability Source: GAO analysis of OMB data.

Percentage default rate or probability

50

40

30

20

10

0 1234567

Years to maturity

Note: OMB default probabilities were calculated from the expected loss
rates that OMB generated for the fiscal year 2004 and fiscal year 2005
budgets. The OMB default probabilities are calculated by removing the
recovery rate adjustments (12 percent for fiscal year 2004 and 9 percent
for fiscal year 2005) from the net default probability tables provided by
OMB, which had recovery rates factored in.

After determining expected default rates, OMB combines the default rates
with an assumption about the recovery rate-the percentage of defaulted
principal and interest that will be recovered over time-to obtain expected
loss rates.37 The assumed recovery rate is a key driver of the expected
loss rates. OMB assumed an across-the-board recovery rate of 17 percent
for the fiscal year 2003 budget-that is, the government was expected to
lose $830 and recover $170 for every $1,000 in defaulted credits. It
assumed

37Expected loss is equal to expected default multiplied by one minus the
recovery rate. Thus, an expected default rate of 15 percent and a recovery
rate of 20 percent would result in an expected loss rate of 12 percent.
(This is expressed mathematically as 15 (1 - .20) =15 (.80) =

                                      12.)

lower recovery rates of 12 percent for the fiscal year 2004 budget and 9
percent for the fiscal year 2005 budget.

OMB's Current Methodology Lowered Expected Loss Rates

OMB's current methodology reduced the loss rates that U.S. credit agencies
are expected to incur on international credits they provide. Between
fiscal years 2002 and 2005, expected loss rates fell across ICRAS risk
categories 1 through 8.38 As shown in figure 4, expected loss rates for
credits of 8-year maturity were, on average, about 58 percent lower on a
present value basis in fiscal year 2005 than 2002 (the last fiscal year in
which OMB used its former approach to develop the loss rates).39 The
largest declines were in risk categories 1 through 5. Expected loss rates
for ICRAS agencies have varied over the credit reform period. (See app.
VIII for information on trends in expected loss rates for ICRAS agencies
between fiscal years 1997 and 2005.)

38Loss expectations rose slightly between fiscal years 2002 and 2005 for
ICRAS categories 9 and 10 and declined slightly for category 11.

39We compared the loss expectations in place for 8-year guarantees at
ICRAS risk ratings 1-8 using Ex-Im Bank's cash flow worksheets and
determined their present value using OMB's credit subsidy calculator. We
selected 8-year credits because this maturity is representative of many
Ex-Im Bank credits. Specific loss expectations per ICRAS category differ
depending on the maturity of the credit, but the general trend and average
reduction over the period were similar for the other maturity bands we
analyzed.

Figure 4: Comparison of OMB's Expected Loss Rates, in Present Value Terms,
for ICRAS Risk Categories 1-8, Fiscal Years 2002�2005

Percentage expected loss rate

40

35

30

25

20

15

10

5

0

Fiscal year

2002

2003

2004

2005

Source: GAO analysis of OMB expected loss rates.

Note: The figure compares loss expectations that were in place in each
fiscal year, in present value terms, for each ICRAS risk category based on
guarantees of 8-year maturity. We based the analysis on 8-year maturities
because this maturity is representative of many Ex-Im Bank credits.

For fiscal year 2003, the first year for which OMB's current methodology
was used to develop expected loss rates, rates declined sharply across
most ICRAS risk categories. Loss rates for fiscal year 2004 rose for
several risk categories, with the biggest change being an increase in the
loss rate for ICRAS risk category 6. According to OMB, the expected loss
rates changed from fiscal year 2003 to 2004 because of updated country
ratings and interest rate data. OMB did not provide more specific
information to explain those changes. Also, the lower recovery rate
assumptions used in

fiscal year 2004 would be expected to push loss rates upward. Expected
loss rates in fiscal year 2005 were generally similar to those in fiscal
year 2004, with slight declines for some risk categories. Although OMB's
model generated lower default rates for fiscal year 2005 than for fiscal
year 2004, a further decrease in its recovery rate assumptions resulted in
little change to expected loss rates in fiscal year 2005.

Current OMB Methodology Has Lowered Ex-Im Bank's Projected Subsidy Costs
and Budgetary Needs

By lowering loss rates for most ICRAS risk categories, OMB's current
methodology has contributed to lower Ex-Im Bank projections of subsidy
costs and, therefore, lower budgetary requirements. Ex-Im Bank's
obligation of budget authority for new subsidy costs declined
significantly for fiscal year 2003, when the current methodology took
effect. In addition, Ex-Im Bank calculated a large downward reestimate of
the subsidy costs of its outstanding portfolio at the end of fiscal year
2002 using the new loss rates. With lower loss rates, Ex-Im Bank's fees
are generally projected to provide greater coverage of expected losses,
fully offsetting losses in some budget categories. Finally, during this
period, Ex-Im Bank modified its approach for calculating loss allowances
in its financial statements. This involved making certain changes to be
more in line with applicable accounting standards and, because of the
changed nature of OMB's loss rates, using different and higher loss rates
than it used in its budget documents to calculate subsidy costs.

Lower OMB Loss Rates Reduce Ex-Im Bank Budget Authority Needed to Cover
Subsidy Costs

Partially because of OMB's lower loss rates, Ex-Im Bank required less
budget authority to cover its lower subsidy costs. Estimates and
obligation of budget authority for subsidy costs are determined by the
amount and risk of business the bank expects to, or does, undertake in a
year, as well as expected loss rates and fees charged to borrowers.
Changes in any one of those factors can alter budget needs. According to
Ex-Im Bank officials, OMB's lower loss rates were a key determinant in the
declines in its subsidy cost estimates, its budget authority obligated for
new subsidy costs, and its 2002 reestimate of subsidy costs.

Ex-Im Bank's requests for budget authority for subsidy costs have dropped
since it began using the lower loss rates to estimate subsidy costs. The
bank's request for subsidy budget authority in fiscal year 2003 was about
30 percent lower than the average of its requests in the previous 5 years,
partly because of lower OMB loss rates and partly because of a substantial

amount of budget authority carried over from the previous fiscal year.40
Ex-Im Bank requested no new subsidy budget authority in fiscal year 2004
but anticipated $460 million in new subsidy cost obligations.41 The amount
of budget authority carried over from previous fiscal years was seen as
sufficient to cover anticipated fiscal year 2004 subsidy costs.42 Ex-Im
Bank requested $126 million for subsidy budget authority in fiscal year
2005, but it anticipated $491 million in obligations for subsidy costs.43
The bank continued to have a significant amount of budget authority
carried over to fund the difference.

In addition, Ex-Im Bank's obligation, or usage, of budget authority for
new subsidy costs dropped when the current methodology took effect, as
shown in figure 5. The bank's obligation of subsidy budget authority had
dropped in fiscal years 2001 and 2002, in part because of reductions in
new financing in those years. Its obligation of budget authority for new
subsidy costs in fiscal year 2003 was about 55 percent lower than in
fiscal year 2002, even though the total amount of its new financing, and
Ex-Im Bank's estimate of its average risk level, in these two fiscal years
was similar.44

40The bank's appropriations in a given fiscal year can be carried over for
up to 3 years if they are not completely obligated. This would happen, for
example, in years when the actual amount and risk ratings of new financing
the bank undertook in a fiscal year were lower than had been anticipated
for that year.

41The figure includes anticipated obligations of $440 million for direct
and guaranteed loan subsidies and $20 million for direct and guaranteed
loan modifications.

42The carryover resulted in part from Ex-Im Bank obligating substantially
less budget authority in fiscal year 2003 than it expected. In its budget
submission, the bank expected to obligate about $655 million for new
subsidy costs and modifications. However, it obligated about $334 million
of the $513 million it was authorized for subsidy costs in fiscal year
2003.

43The bank anticipated obligations of $471 million for direct and
guaranteed loan subsidies and $20 million for direct and guaranteed loan
modifications.

44In 2002, Ex-Im Bank authorized about $10.1 billion in new financing,
with an average ICRAS risk weight of 4.9; its obligation of subsidy budget
authority that year was about $738 million. In fiscal year 2003, Ex-Im
Bank authorized about $10.5 billion in new financing, with an average
ICRAS risk weight of 5.0; its obligation of subsidy budget authority that
year was about $334 million.

Figure 5: Ex-Im Bank Obligation of Budget Authority for New Subsidy Costs,
Fiscal Years 1992-2003

Dollars in millions 1,000

900

800

700

600

500

400

300

200

100 0

          1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003

Fiscal year Source: GAO analysis of Ex-Im Bank Financial Highlights.

Note: Figure excludes subsidy budget authority obligated for purposes of
tied-aid (government-togovernment concessional financing of public sector
capital projects in developing countries that is linked to the procurement
of goods and services from the donor country).

According to Ex-Im Bank officials, OMB's lower loss rates also contributed
to a significant downward reestimate of the subsidy costs of the bank's
outstanding credits, based on its first subsidy cost reestimate that used
the lower rates.45 At the end of fiscal year 2002, using the OMB loss
rates for

45When calculating its reestimates, Ex-Im Bank uses the most current OMB
loss rates available. According to Ex-Im Bank officials, its reestimate in
fiscal year 2002 was calculated using fiscal year 2004 loss rates. The
officials said that the fiscal year 2003 loss rates were not used for any
reestimate calculations.

fiscal year 2004, Ex-Im Bank calculated a net downward reestimate of about
$2.7 billion, significantly lowering its estimated subsidy costs for
outstanding credits.46 Downward reestimates on long-term guarantees
represented about 72 percent of the reestimate. About 63 percent of the
reestimate was calculated on financing extended between fiscal years 1997
and 2001, much of which has likely not yet matured.47

With Lower Loss Rates, Ex-Im Bank Fees Are Projected to Provide Greater
Coverage of Losses

With the decline in OMB's loss rates, Ex-Im Bank's exposure fees are
projected to generally provide greater coverage of its expected losses.48
The determination of the relationship between exposure fees and expected
losses and, thus, the calculation of budget subsidy cost, depends on the
risk rating for specific Ex-Im Bank transactions. Ex-Im Bank generally
sets its exposure fees at, or in the case of some corporate transactions
slightly above, the minimum level required by an agreement among certain
OECD member countries.49

This agreement among OECD countries was designed to increase transparency
and provide common benchmarks for ECA exposure fees, thereby reducing fee
competition among exporters. Participating ECAs may charge fees above the
OECD minimum if they do not view the fees as sufficient to cover their
expected losses on a given transaction, but they are expected to charge at
least the minimum. For private sector transactions,

46The total downward reestimate that Ex-Im Bank calculated in fiscal year
2002 was about $3.5 billion, of which $2.7 billion was reestimated subsidy
cost and $0.8 billion was interest cost. At the end of fiscal year 2003,
Ex-Im Bank calculated a net downward reestimate of about $1.9 billion, of
which about $1.4 billion was reestimated subsidy costs and about $0.5
billion was interest costs.

47Percentages on the 2002 reestimates were calculated on subsidy and
interest costs combined because information on the trends by cohort was
available only in this format.

48Exposure fees are fees that Ex-Im Bank charges borrowers to cover the
risk that the transaction will not be repaid. These fees vary depending on
the risk and tenor of the credit being offered. Ex-Im Bank also charges
other fees, including application processing fees and commitment fees.

49This agreement was among Participants to the Arrangement on Officially
Supported Export Credits. Participants to the Arrangement are Australia,
Canada, the members of the European Community, Japan, the Republic of
Korea, New Zealand, Norway, Switzerland, and the United States. The fee
agreement, sometimes called the Knaepen Package after the Belgian official
instrumental in its formation, was concluded in 1997 and took effect in
1999. The fees in the agreement result from a political agreement, an
averaging of fees in place in 1996 across certain export credit agencies.

participating ECAs that we spoke with often charge fees above the OECD
minimum fees. (See app. II for additional information on the OECD minimum
fee determination process.)

Using fiscal year 2005 expected loss rates, Ex-Im Bank exposure fees at
the OECD minimum fee level would be projected to fully cover expected
losses in ICRAS categories 1-5 in certain cases (see fig. 6). In
comparison, using fiscal year 2002 expected loss rates, Ex-Im Bank
exposure fees at the OECD minimum fee level were projected to cover
expected losses only for ICRAS category 1.

Figure 6: Comparison of Ex-Im Bank Exposure Feesa and Expected Loss Rates
by ICRAS Category, Fiscal Years 2002 and 2005

Percentage expected loss 40

35

30

25

20

15

10

5

0

FY 2002 FY 2005 Ex-Im Bank exposure fee based on OECD minimum premia
benchmarks Sources: GAO analysis of OMB expected loss rates and Ex-Im Bank
exposure fee based on OECD minimum premia benchmarks.

aFigure compares Ex-Im Bank exposure fees at the minimum OECD fee level
with GAO's analysis of expected loss for credits of 8-year maturity in
fiscal years 2002 and 2005 (see fig. 4).

The degree to which Ex-Im Bank's exposure fees are projected to cover its
expected losses may differ from this illustration, depending on the type
of borrower or transaction. For example, when Ex-Im Bank assigns a
corporate borrower a higher risk rating than that of the country where the
borrower is located, the bank may incur subsidy costs in more risk
categories or may incur larger subsidy costs for corporate borrowers rated
in categories 6 through 8.50 This is because Ex-Im Bank charges fees for
corporate transactions that are close to the OECD minimum fee for the
country in which the corporate borrower is located, even when the
transaction has a higher (riskier) rating than the country. In addition,
the OECD guidance does not apply to some transactions, notably aircraft
financing.

In generally setting exposure fees at or near the OECD minimum level,
Ex-Im Bank charges fees that are among the lowest of ECAs. Ex-Im Bank's
low pricing relative to other ECAs has been noted for some time. According
to U.S. and OECD officials, whereas Ex-Im Bank previously appeared to face
some pressure to charge higher fees because of its budget costs (and
appeared to support raising the minimum OECD fees as well), the lower
budgetary costs of Ex-Im Bank's activities have lessened this pressure.51

Ex-Im Bank Modified Its Method for Determining Financial Statement Loss
Estimates, Generally Using Higher Loss Rates Than for Budget Calculations

Beginning with its 2002 financial statements, Ex-Im Bank modified its
approach for calculating loss allowances, which involved segmenting its
portfolio in line with applicable accounting standards and diverging from
its former practice of using OMB loss rates to calculate these allowances.
Because Ex-Im Bank prepares its financial statements according to private
sector, rather than federal, accounting principles, there has always been
some difference between the bank's subsidy cost and loss allowance
estimates. This is because of differences in the treatment of fee income

50For corporate borrowers that Ex-Im Bank rates as riskier than the OECD
sovereign rating for the country where the corporation is located, the
bank charges fees that are higher than the OECD fee by 10 percent
increments for each difference in rating level. Thus, if Ex-Im Bank rated
a corporation in Country A at ICRAS category 5 when Country A is rated at
ICRAS category 3, the bank would charge that corporation an exposure fee
20 percent higher than the OECD fee corresponding to ICRAS category 3.
However, because that fee would not cover expected loss in ICRAS category
5, Ex-Im Bank would incur subsidy costs in this example.

51A foreign government official noted that Ex-Im Bank exhibits a greater
degree of transparency than other members of the export credit group by
explicitly disclosing the subsidy cost of its export credit activities in
its budget documents.

between private sector and federal accounting approaches.52 While Ex-Im
Bank is not required to use OMB loss rates when calculating financial
statement loss allowances, Ex-Im Bank officials said that they had
historically chosen to do so in order to link the loss estimates prepared
for budget purposes with the financial statement loss allowances. However,
in its 2002 financial statement, Ex-Im Bank began applying higher loss
rates than OMB's loss rates to most of its portfolio. Ex-Im Bank officials
said that with the modification, its approaches to calculating subsidy
costs and loss allowances differ not only in fee treatment but also in
their expectations of loss.

According to Ex-Im Bank officials, the bank modified its financial
statement loss allowance methodology for two reasons. First, Ex-Im Bank
discussed with its new auditors, Deloitte & Touche, the bank's approach
for accounting for guarantees and insurance, which comprise a majority of
the bank's portfolio.53 They determined that relevant accounting standards
suggested that it would be appropriate to record these credits at their
fair market value. This called for using different loss rates than those
derived using OMB's current methodology, which focuses on credit loss.54
Second, the bank determined that it should value its impaired credits in a
manner more consistent with relevant accounting standards.55 Deloitte &
Touche observed that Ex-Im Bank had not historically separated its
portfolio into impaired and unimpaired groupings in accordance with
accounting guidance, even though a significant portion of these loans and
claims were likely impaired. Total loans and claims represented, on
average, about 24 percent of the bank's total exposure during fiscal years
1999-2003.

In addition, when initially estimating its 2002 loss allowances using its
former approach and OMB's fiscal year 2003 loss rates, Ex-Im Bank
determined that its allowances would have dropped substantially, from

52Private sector accounting does not recognize future fee income before it
is received. In contrast, for budget purposes, the present value of future
fee income is recognized as an offset to expected losses when a loan or
guarantee is made.

532002 was the first year Deloitte & Touche served as Ex-Im Bank's
external auditor.

54FASB Interpretation Number 45 addresses Guarantors' Accounting and
Disclosure Requirements for Guarantees, Including Indirect Guarantees of
Indebtedness of Others, an interpretation of FASB Statements Nos. 5, 57,
and 107 and rescission of FASB interpretation No. 34.

55Ex-Im Bank officials cited Statement of Financial Accounting Standard
114, which addresses Accounting by Creditors for Impairment of Loan.

about $10 billion for 2001 to about $6 billion for 2002, a decrease of
about 40 percent. Because of the size of the reduction and the importance
of loss allowances as an overall reflection of an institution's expected
loss from year to year, the bank's auditor identified this as a key area
to be reviewed.

Ex-Im Bank's approach for calculating its loss allowances, beginning with
the 2002 financial statement, used different loss rate methodologies for
different parts of its portfolio and distinguished between impaired56 and
unimpaired credits.57 To determine the loss allowances for its impaired
loans and claims and for all of its loan guarantees and medium- and
longterm insurance, Ex-Im Bank applied higher loss rates than those that
were used in 2001. These higher rates were used to calculate about 95
percent of the 2002 loss allowance. Ex-Im Bank had asked OMB to provide
these higher rates using its former, spread-based methodology.58 Ex-Im
Bank officials stated that the spread-based loss rates were more
appropriate for its outstanding guarantees and insurance because they
provided a more market-based valuation that was better suited to a fair
value presentation. To determine the 2002 loss allowances for its
unimpaired loans and claims, Ex-Im Bank applied OMB's expected loss rates
for the fiscal year 2004 budget.59 These rates were generally lower than
the rates used to calculate the loss allowance in 2001. For 2002, the
bank's loss allowances were about

56Ex-Im Bank defined as impaired any loans or claims that are 90 days or
more in arrears, that are rated in ICRAS categories 9-11, or that have
been rescheduled. The bank determined that about 34 percent of the loans
and about 95 percent of the claims it reported in its 2002 financial
statement were impaired. For fiscal year 2003, about 23 percent of its
loans and 95 percent of its claims were impaired.

57Ex-Im Bank discloses detailed information about its loss allowance
calculation, including the different loss rates it uses, in its Annual
Portfolio Review, which is compiled by its Portfolio Management and Review
Division.

58Ex-Im Bank also applied different loss rates depending on whether the
exposures were outstanding or undisbursed. To discount the effect of
exposure of cancellations and suspension of disbursements, Ex-Im Bank set
the loss percentage for each ICRAS category of undisbursed exposure 15
percent lower than the loss percentage for outstanding exposures.

59Usually, the loss rates that Ex-Im Bank applies to its financial
statements are the rates that will be in effect in the coming fiscal year.
According to an Ex-Im Bank official, fiscal year 2003 expected loss rates
were not used in the 2002 financial statement because OMB made the fiscal
year 2004 rates available sooner than expected.

6 percent higher than their 2001 level. For 2003, loss allowances were
about 4 percent lower than their 2002 level.60

Loss Estimation Involves Challenges and OMB Methodology Is Not Transparent

OMB's methodology for estimating the expected loss rates for international
credits provided by U.S. agencies involves challenges, and it is not
transparent. Assessing the risk of such credit activity, particularly in
developing countries, is inherently difficult. Corporate default data
similar to those used by OMB are also used by other financial institutions
to assess risk, because of the data's broad coverage and limitations in
other data sources. However, historically, the data have been based
largely on the default experiences of U.S. firms, and the data's
historical coverage of developing countries has been limited. In addition,
more recent Moody's data than were used in estimating OMB's model show
higher defaults in some risk categories. In choosing this data to predict
default, OMB analyzed Ex-Im Bank defaults over a somewhat narrow period.
In addition, while OMB has assumed increasingly lower recovery rates since
implementing its method, its basis for the recovery rates and changes in
them has not been transparent. Finally, despite its complexity and the
changes it implied, OMB developed the current methodology independently
and provided ICRAS agencies with limited information about the
methodology.

Assessing ECA Financing Risk is Difficult, and Data Used by OMB May Have
Inherent Limitations for Predicting Ex-Im Bank Risk

Assessing ECA financing risk presents data challenges. Available
indicators of default risk, including certain financial institutions' own
financing histories, often have limitations. Historical data on corporate
bond defaults, while used by many institutions, may also have inherent
limitations for assessing risk in developing countries, because these data
have historically been based primarily on corporations in higher-income
countries. In addition, corporate default data now show higher defaults in
certain higher-risk categories than the data OMB used. OMB's analysis
showing comparability between those data and Ex-Im Bank default

60Ex-Im Bank's financial statement loss allowances have averaged about 18
percent of its total exposure since fiscal year 1999. Ex-Im Bank did not
establish any allowances for credit losses on its loans and loan
guarantees through fiscal year 1988, suggesting that no expected losses
were associated with any of these credits. As the bank's independent
auditor at this time, GAO expressed adverse opinions about the bank's
financial statements for fiscal years 1983-1988, noting that the financial
statements were not fairly presented in accordance with GAAP. Ex-Im Bank
established financial statement loss allowances for the first time in
1989, in the amount of $4.8 billion.

experience was based on Ex-Im Bank credits primarily from a relatively
narrow period in the 1990s and did not include other ICRAS agencies. OMB
representatives, in providing oral technical comments on a draft of this
report, said that they inquired about available default data at other
ICRAS agencies but were unable to obtain it.61 OMB's staff paper noted the
desirability of adding data from other agencies to its analysis in the
future.

Assessing ECA Financing Risk Is Data limitations and changing environments
present challenges for

Difficult

estimating the risk of ECA financing, according to experts and officials
with whom we spoke. Some noted that ECA risk may differ from private bank
risk, in that ECAs may be more exposed to emerging markets and may have
less diversified portfolios, in part because of concentrations of exposure
to particular industries. Officials said that many ECAs incurred large
losses during the 1980s debt crisis, and some did so in the 1990s during
the Asian financial crisis and other instances of sovereign default.
Moreover, the move among some ECAs, including Ex-Im Bank, toward extending
more credit to corporate or other nonsovereign borrowers, rather than
primarily providing financing to sovereign governments, adds further
complexity to estimating risk. According to several ECA officials,
corporate activity may involve different risks, which include potentially
greater difficulty in recovering assets in cases of default.

Some ECAs and other financial institutions lack data on their own
financing that are of sufficient historical coverage and reliability for
predicting the risk of future financing activities. In addition, a lack of
risk ratings for financing in earlier decades can complicate the use of
available historical data. Historical data on the default experience of
sovereign bond issuers might be useful in estimating ECA credit risk, and
ratings agencies now publish such data. However, according to several
experts, the limited risk rating history of sovereign bond issuers is a
significant limitation to relying on this data to assess risk in
developing countries. Almost no

61Assessing the availability of default data at other ICRAS agencies was
beyond the scope of this review.

developing country sovereign bond issuers have ratings histories that
begin before the early 1990s.62

Corporate Bond Default Data Corporate bond default rates from nationally
recognized rating agencies

Are Widely Used but Lack Broad are widely used by financial institutions
in assessing risk but may have

Historical Coverage of certain inherent limitations for predicting
defaults in developing countries.

Developing Countries	Institutions use the data because of the large number
of firms and long historical coverage in the rating agencies' databases.
However, while international corporations are now well represented in
these data, the data historically have included primarily U.S. firms.63
For data with international coverage, the coverage has historically been
largely of high-income country borrowers. One study of a major rating
agency database found, for example, that 94 percent of the nonbank firms
rated were in high-income countries, 5 percent were in upper-middle-income
countries, and 2 percent were in lower-middle-income countries.64 The
data's more limited historical coverage of developing country default
experiences may limit the predictive value of the data for such countries,
according to some officials. Officials from one institution said that
although they used corporate bond data in determining expected default
rates, whether countries were in emerging markets was a consideration in
their adjustments of the default rates to reflect their own performance
expectations.

62For example, in 1985, Moody's and Standard & Poor's rated a total of 15
sovereigns, with 14 of the countries rated in categories corresponding to
ICRAS category 1 and 1 corresponding to ICRAS category 5. In 1991, they
rated a total of 34 countries, with 21 of the countries rated in
categories corresponding to ICRAS category 1, 6 rated in categories
corresponding to ICRAS category 2, and 7 rated in categories corresponding
to ICRAS categories below 2. By 1999, these rating agencies rated a total
of 91 countries, with 58 rated in categories below those corresponding to
ICRAS category 2. (These figures use an average of the two agencies'
ratings, in cases where they rated a country differently in a given year.)

In addition to the limited coverage of ratings, experts have noted that
countries historically have often given preferential treatment to payments
on their bonds compared with their loans because bonds represented a
relatively small share of a country's international debt.

63For example, in 1980, a very small percentage of bond issuers rated by
Moody's were located outside the United States; the non-U.S. share grew to
about 18 percent by 1990 and 40 percent by 2000.

64The study examined Standard & Poor's data on rated corporations as of
1999. See Giovanni Ferri, Li-Gang Liu, and Giovanni Majnoni,"The Role of
Rating Agency Assessments in Less Developed Countries: Impact of the
Proposed Basel Guidelines," Journal of Banking and Finance 25 (2001):
115-148.

More Recent Moody's Data Show Higher Defaults in Some Risk Categories

OMB Used Ex-Im Bank Data That Primarily Covered a Limited Period to
Establish Comparability with Corporate Data

More recent Moody's bond default rates are higher for some higher-risk
categories than the Moody's data for 1983-1999 that OMB's model uses to
estimate default rates. The more recent data show higher default rates for
risk levels that correspond to ICRAS categories 7 and 8, the highest risk
categories in which Ex-Im Bank undertakes new business. For example, for
fiscal year 2005, OMB's model predicted a default rate for ICRAS category
8, assuming a maturity of 8 years, of 41 percent. The Moody's default rate
for 1983-1999 was 48 percent, whereas the rate was 52 percent for
19832001, and 58 percent for 1983-2003. Based on our review of available
information on OMB's default model, the model would be expected to
generate higher default rates for these categories if these more current
Moody's data were used.

In deciding to use corporate default data to predict U.S. international
credit agencies' defaults, OMB compared data on Ex-Im Bank historical
defaults, primarily from a limited period, with corporate default rates.
Among ICRAS agencies, Ex-Im Bank generally extends the largest portion of
the U.S. government's new foreign credit exposure each year.65 OMB did not
compare other ICRAS agencies' defaults with the corporate data. OMB
representatives, in providing oral technical comments on a draft of this
report, said they inquired at other ICRAS agencies about default data but
were unable to obtain additional data. OMB recognized in its staff paper
the value of adding other agencies' data to its analysis in the future.

The Ex-Im Bank credits that OMB analyzed were primarily from a relatively
narrow historical period in the 1990s. OMB examined the default
probabilities of certain Ex-Im Bank transactions, sorted by risk rating,
and concluded that Ex-Im Bank default rates were generally somewhat lower

65Since credit reform, the value of Ex-Im Bank's annual loans and loan
guarantees have been the largest among ICRAS agencies, followed by the
Commodity Credit Corporation and the Overseas Private Investment
Corporation. On a total exposure basis, at the end of fiscal year 2002,
Ex-Im Bank represented 40 percent of the U.S. government's credit exposure
to sovereign and other official foreign borrowers; the Agency for
International Development and the Departments of Agriculture and Defense
represented most of the remainder. At that time, Ex-Im Bank also
represented about 70 percent of the U.S. government's credit exposure to
private foreign borrowers, with the remainder held primarily by the
Overseas Investment Protection Corporation and the Department of
Agriculture.

than those of corporate bonds across comparable rating categories.66
However, a comparison based primarily on lending activity over a
relatively short time frame may not be representative of Ex-Im Bank's
overall default risk. In addition, according to several experts and
officials, data that reflected only the international business climate of
the 1990s would not be representative of the risk of international
lending.

For this comparison, OMB used data from an Ex-Im Bank database covering
guarantees and medium-term insurance transactions from fiscal years
1985-1999. This data set did not include loans, which comprised a
significant part of Ex-Im Bank financing through the early 1980s, and
which experienced substantial defaults during an international debt crisis
that began in the early 1980s.67 While we could not determine the specific
data that OMB analyzed,68 our analysis of the 1985-1999 database indicated
that the majority of observations in the overall database, and a strong
majority

66OMB initially calculated default probabilities directly for different
risk categories. Because of the relatively small number of observations in
the database, the patterns shown were somewhat erratic, with higher
percentages of defaults in some lower-risk categories than in higher-risk
categories. OMB then used a statistical model to smooth the patterns of
historical defaults across ratings categories for ICRAS categories 1-6.

67According to Ex-Im Bank officials, in 2000, they provided OMB a database
of the bank's guarantees and medium-and long-term insurance financing
activities covering fiscal years 1985-1999, which Ex-Im Bank had created
to provide information to a private bank with which they were considering
joint financing. According to Ex-Im Bank officials, they did not include
loans in the creation of that data set in 1999 because loans had become
less important in the bank's financing. Ex-Im Bank officials told us that
the reliability of this data is considerably greater for transactions
initiated in 1996 or later. Beginning in 1996, the key component databases
were updated to increase automatic data entry and verification.

68In a written response to our questions, OMB provided information
indicating that its analysis was based on data from 1993 and later, but
OMB staff stated subsequently that the analysis had used all observations
in the data set. Within the data set, credits prior to fiscal year 1992
lack ICRAS risk ratings. OMB staff stated that they assigned ratings by
using risk ratings from private rating agencies. However, we determined
that very few countries below the highest rating categories were rated by
private rating agencies before 1992. This is discussed in footnote 62.

of observations for which risk ratings were available, were from the
mid-to late-1990s.69

              Recovery Rate Assumptions Have Not Been Transparent

The basis for OMB's recovery rate assumptions and the changes over time
has not been transparent. During our audit work, OMB did not respond to
questions about the specific basis for its recovery rate assumptions of 17
and 12 percent, respectively, for fiscal years 2003-2004. OMB further
reduced assumed recovery rates to 9 percent for fiscal year 2005. In
discussing recovery rate assumptions during the audit work, OMB cited its
staff paper, which contained a recovery rate of 20 percent that OMB said
was based generally on the ratio of aggregate recoveries to aggregate
claims in Ex-Im Bank historical data. However, in discussions on a draft
of this report, OMB representatives said that the market price of credits
with the lowest ICRAS rating (category 11) was the predominant basis for
recovery rates, although they did initially also consider data on Ex-Im
Bank recoveries. OMB representatives said using the market price of the
lowestrated credits is based on the assumption that this value represents
the most the U.S. government would recover in the event of a default. They
provided information to show that changes in OMB's calculation of market
prices of these credits accounted for drops in the recovery rate
assumptions over time.70 Changes in OMB's calculation of these prices
resulted in part from technical comments by Treasury officials.

Our analysis of the 1985-1999 Ex-Im Bank data indicates that the ratio of
aggregate recoveries to aggregate claims in that database is about 19

69We obtained the 1985-1999 data set from Ex-Im Bank and examined the
distribution of transactions across the period covered. We determined
that, depending on how one defines the unit of analysis, from two-thirds
to over three-fourths of the observations in the data set for which
Moody's Investors Service, Standard & Poor's, or ICRAS country ratings
were available were from 1994-1999. We also determined that, irrespective
of which observations had ratings associated with them, between slightly
more than half and two-thirds of them were from 1994 or later. Between 85
and 91 percent of all the observations in the database were from 1990 or
later.

70Footnote 33 provides more information on how OMB uses market prices in
calculating expected losses for ICRAS categories 9-11.

percent.71 Recovery rates that were based on aggregate data over a limited
time period would tend to underrepresent actual recoveries because of the
limited period for recoveries to be observed, especially those associated
with defaults occurring at the end of the period.

According to financial institution officials and rating agency analysis,
recovery rates tend to vary by borrower type and risk and can fluctuate
cyclically. OMB's recovery rates assumptions appear to be conservative
compared with recovery rates assumed by other financial institutions.
Institutions we talked to generally assumed higher recovery rates than
OMB, and some tailored their recovery rate assumptions according to the
type of borrower. According to rating agency analysis, recovery rates are
generally lower for riskier credits and fall during periods when defaults
are higher.

If recovery rates assumed by OMB are lower than likely Ex-Im Bank
recoveries, they will offset, to some degree, lower expected defaults in
the calculation of expected losses. Because expected losses are calculated
by combining expected default and expected recovery rates, an
unrealistically low recovery rate would necessarily offset an
unrealistically low expected default rate, within certain ranges.72

71We analyzed the Ex-Im Bank data sets primarily at the aggregate level.
Recovery amounts in this data set represent recoveries received directly
by Ex-Im Bank and do not include payments to the U.S. government through
reschedulings of sovereign credits. Ex-Im Bank provided us a second data
set that updated the 1985-1999 data through fiscal year 2001. Our analysis
showed that the ratio of aggregate recoveries to aggregate claims in the
second data set is about 35 percent. Ex-Im Bank officials said that, when
recoveries through reschedulings are included, their total recovery rates
are higher. Our analysis showed that, when recoveries through
reschedulings are included, total recovery rates were about 26 percent for
1985-1999 period and about 43 percent for the 1985-2001 period.

72For example, in a hypothetical situation where the true default
probability was known to be 20 percent and the true recovery rate was 30
percent (corresponding to a loss rate of 70 percent), the true expected
loss rate would be 14.0 percent (20 x .70=14.0). However, an overly low
default rate of 15 percent combined with an overly low recovery rate of 10
percent (corresponding to a loss rate of 90 percent) would yield a similar
expected loss rate, 13.5 percent (15 x .90=13.5).

OMB Developed Method Independently and Provided Agencies Limited
Information

Because of OMB's unique role in developing the loss rates that ICRAS
agencies use to calculate subsidy costs, these agencies rely on OMB to
comply with credit reform requirements that address the agencies'
responsibilities for assuring the reliability of their subsidy cost
estimates. Despite the complexity of the current methodology and its
implications for ICRAS agencies' subsidy costs, OMB developed the current
methodology predominantly on its own, receiving some input from one ICRAS
agency. Some ICRAS officials said that OMB provided agencies with limited
information about the methodology's basis or structure.

Credit reform guidance on developing credit subsidy estimates addresses
the procedures and internal controls that agencies should have in place to
ensure that their estimates are reliable.73 It states that any changes in
factors and key assumptions, such as default and recovery rates, should be
fully explained, supported, and documented. The purpose of thorough
documentation is to enable independent parties to perform the same steps
and replicate the same results with little or no outside explanation or
assistance.

OMB representatives said that the current methodology was reviewed within
OMB and circulated among the ICRAS agencies, although several ICRAS agency
officials told us OMB had provided them limited information. According to
these officials, OMB presented its methodology as an essentially completed
approach and held several meetings during 2001 and early 2002 to discuss
it. Officials who received information and attended certain meetings told
us that it was difficult to understand or evaluate the methodology based
on the information provided. For example, prior to one meeting, OMB
circulated a two-page discussion paper that discussed OMB's rationale for
adopting the current methodology and generally described its approach and
a technical appendix to a staff paper that contained numerous equations
describing a theoretical model.

73Federal Accounting Standards Advisory Board, Federal Financial
Accounting and Auditing Technical Release 6-Preparing Estimates for Direct
Loan and Loan Guarantee Subsidies under the Federal Credit Reform Act,
Amendments to Technical Release 3: Preparing and Auditing Direct Loan and
Loan Guarantee Subsidies Under the Federal Credit Reform Act (Washington,
D.C.: January 2004). This guidance was developed by the Accounting and
Auditing Policy Committee, a permanent committee of the Federal Accounting
Standards Advisory Board. The committee was organized by OMB, GAO, the
Department of Treasury, the Chief Financial Officers' Council, and the
President's Council on Integrity and Efficiency as a body to research
accounting and auditing issues requiring guidance.

However, OMB representatives told us that some of the equations in this
appendix were not actually used in the methodology while other equations
not contained in the appendix were used. At one meeting, according to a
Department of Agriculture economist, OMB provided only the expected loss
rates for fiscal year 2003 and a graph that predicted a decline in Ex-Im
Bank subsidy rates. This official said it was not possible to understand
the methodology by only examining its results. She said that although she
and other ICRAS agencies representatives posed various questions about the
method's underlying data and assumptions, OMB representatives did not
provide substantive responses and stated that the method was too complex
to explain. OMB representatives said the methodology was not reviewed
outside the U.S. government.

After presenting the methodology, OMB received comments on the methodology
from at least one ICRAS agency. Treasury officials told us that when they
examined the proposed expected loss rates for fiscal year 2003, they
objected to the substantially lower loss rates for the riskiest countries,
those in ICRAS categories 9 through 11; asked to see the underlying data
used; and raised methodological issues regarding how those rates were
calculated.74 The Treasury officials said that while they had some
questions about the expected loss rates for other ICRAS categories, they
focused their attention on the treatment of the riskiest countries. The
reason, they said, was that planned drops in expected loss estimates for
these countries would sharply increase the cost to the United States of
forgiving existing debt, such as through international agreements to
forgive the debt of highly indebted poor countries.75 According to
Treasury officials, OMB revised its approach for estimating expected
losses in ICRAS categories 9 through 11, which resulted in loss rates that
were not significantly changed from those in effect before fiscal year
2003.

We found that some financial institutions used outside experts or
consultants in developing their loss estimation methodologies. Some also
described procedures that exist to ensure their methodology's ongoing
objectivity and reliability. For example, other government agencies, audit

74Treasury officials determined, for example, that some credits that were
being used to determine the market value and implicitly the riskiness of
lower-rated country data were backed with collateral, which would be
expected to result in lower risk and higher prices.

75When the U.S. government forgives a country's debt, the budgetary cost
of the debt relief is determined by the estimated value of the debt under
credit reform terms. Thus, the lower the estimated value of the debt, the
lower the budgetary cost of debt forgiveness.

organizations, and outside experts have been involved in developing or
reviewing the methodologies of two foreign ECAs that we contacted. Also,
regulatory bodies, audit organizations, and internal risk management
groups are involved in overseeing bank loss estimation methodologies.

Conclusions	In passing the Federal Credit Reform Act in 1990, Congress
required agencies to develop reasonable estimates about the long-term cost
to the government of federal credit programs, to ensure a sound basis for
decisions regarding program budgets. For international credit agencies
such as Ex-Im Bank, which finances activities in relatively risky markets,
predicting long-term costs and determining appropriate budget subsidy
amounts is especially challenging. Because of the importance of reasonable
program cost estimates under credit reform, such estimates need to be made
with appropriate data and using appropriate analytical techniques. While
ICRAS agency subsidy costs have several determinants, including the
particular risk ratings assigned to different borrowers, OMB's directions
to ICRAS agencies regarding loss rates across risk levels are an important
element of estimating subsidy costs.

OMB's shift to using historical corporate default data in its methodology
for estimating loss rates of ICRAS agency activities has some basis, given
the practices of other financial institutions and limitations in available
historical data. However, the predictive value of those corporate default
data for the financing undertaken by Ex-Im Bank or other ICRAS agencies
has not yet been established. Obtaining additional information on
agencies' default and repayment experiences over time will allow better
assessments of the suitability of using data such as corporate bond
default rates.

The lack of transparency of OMB's current loss rate methodology raises
questions about how it determines expected loss rates. Because of this
lack of transparency, combined with the method's complexity, the multiple
ICRAS agencies that use the loss rates have incomplete information about
how those rates are determined and what factors are driving changes over
time. OMB's unique role in setting ICRAS agency loss rates suggests that
greater transparency would be appropriate. In addition, other credit
reform tools that multiple credit agencies use to calculate subsidy costs,
such as OMB's credit subsidy calculator, have been audited to assure users
about their accuracy. Independent review of OMB's methodology would
provide similar assurance about the reliability of the loss rates and the
subsidy costs developed from these rates, and could help facilitate ICRAS
agency financial statement audits.

Recommendations for Executive Action

To improve the transparency of the subsidy cost estimation process and
help ensure the validity of estimates over time, we recommend that the
Director of the Office of Management and Budget take the following five
actions:

o 	Provide ICRAS agencies and Congress a technical description of OMB's
expected loss methodology, including the default model, the key
assumptions OMB made, and the data it used.

o 	Provide similar information in the event of significant changes in its
method of calculating expected loss rates.

o 	Ensure that data from nonagency sources-for example, rating agencies'
corporate default data, which are used to estimate expected loss rates-be
updated as appropriate.

o 	Request from Ex-Im Bank and other U.S. international lending agencies
the most complete and reliable data on their default and repayment
histories and periodically obtain updated information, so that the
validity of the data on which the current methodology is based can be
assessed as sufficient agency data are available.

o 	Arrange for independent methodological review of OMB's expected loss
rate model and assumptions and document that review.

Agency Comments and Our Evaluation

We provided a draft of this report for formal comment to the Director,
Office of Management and Budget; the Chairman, Export-Import Bank; the
Secretary of the Treasury; the Chairman, Board of Governors of the Federal
Reserve System; the Comptroller of the Currency; and the Chairman, Federal
Deposit Insurance Corporation. We also provided a copy of the draft report
for technical review to the Chairman, Securities and Exchange Commission,
and officials at the Foreign Agricultural Service of the Department of
Agriculture. OMB provided written comments on the draft report, which are
reprinted in appendix IX. OMB, Ex-Im Bank, the Comptroller of the
Currency, and the Securities and Exchange Commission provided technical
comments, which we incorporated as appropriate. Other agencies reviewed
the report but had no comments. We also obtained technical comments from
bank and foreign ECA officials on our descriptions of their practices.

OMB generally agreed to implement the report's recommendations to make
more information available on its expected loss methodology, update the
nonagency data used in the model, obtain additional agency default data
over time, and obtain technical review. OMB also expressed concern about
the report's statement that OMB's method for determining loss rates was
not transparent, observing that our report generally describes the method.
We believe that, while we do present in this report a substantial amount
of information on OMB's loss methodology, obtaining that information
required considerable resources and effort with certain information
provided only during the agency comment period despite repeated inquiries
by GAO, and that similar information should be more readily available to
affected agencies and Congress on an ongoing basis.

We are sending copies of this report to appropriate Congressional
Committees. We are also sending copies of this report to the Director,
Office of Management and Budget; the Chairman, Export-Import Bank; the
Secretary of the Treasury; the Chairman, Board of Governors of the Federal
Reserve System; the Comptroller of the Currency; the Chairman, Federal
Deposit Insurance Corporation; the Chairman, Securities and Exchange
Commission; and the Administrator, Foreign Agricultural Service of the
Department of Agriculture. We also will make copies available to others
upon request. In addition, the report will be available at no charge on
the
GAO Web site at http://www.gao.gov.

If you or your staff have any questions about this report, please contact
me
on (202) 512-4346. Additional GAO contacts and staff acknowledgments are
listed in appendix X.

Loren Yager, Director
International Affairs and Trade Issues

Appendix I

                       Objectives, Scope, and Methodology

The Export-Import Bank Reauthorization Act of 2002 directed GAO to report
to the House of Representatives Committee on Financial Services and the
Senate Committee on Banking, Housing and Urban Affairs on the reserve
practices of the Export-Import Bank (Ex-Im Bank) as compared with the
reserve practices of private banks and foreign export credit agencies
(ECA). The committees were specifically interested in Ex-Im Bank's method
for estimating the subsidy costs of its financial activities for budgetary
purposes in accordance with the Federal Credit Reform Act of 1990. Ex-Im
Bank subsidy costs are determined, in part, on the basis of a methodology
established by the Office of Management and Budget (OMB); OMB's
methodology changed substantially in fiscal year 2003.

In response to the mandate, we agreed to (1) describe OMB's current and
former methodologies for estimating expected loss rates for international
credits and the rationale for the recent revisions, (2) determine the
impacts of the current OMB methodology on Ex-Im Bank, and (3) assess the
current methodology and the process by which it was developed. We also
agreed to provide information on the reserve practices of foreign ECAs and
commercial banks.

To describe OMB's current and former methodologies for estimating expected
loss rates for U.S. credit agencies' international credit and the
rationale for the recent revisions, we reviewed OMB descriptions of the
methodologies and discussed the rationale for the changes to the
methodology with OMB staff and Ex-Im Bank, Treasury, and Congressional
Budget Office officials. We also reviewed finance literature that OMB
cited as a basis for modifying its approach, as well as related
literature, and examined Ex-Im Bank information about trends in its
subsidy cost reestimates (discussed later). Our description of the former
methodology is also based on prior GAO work, on OMB memoranda to agencies
that participate in the Interagency Country Risk Assessment System (ICRAS)
announcing the risk premiums or expected loss rates to be used in
preparing budget estimates for upcoming fiscal years, and on our analysis
of expected loss rates for fiscal years 1997-2002, as described later. We
generally describe how ICRAS risk ratings are established and how Ex-Im
Bank rates private borrowers, and while we recognize that the assignment
of risk ratings is an important element in the overall reasonableness of
expected loss estimates, evaluating the reasonableness of the risk ratings
process and of specific ratings was beyond this scope of this engagement.

To describe the current methodology, we examined OMB's written
descriptions of its methodology, posed specific questions to OMB staff on

Appendix I
Objectives, Scope, and Methodology

several occasions through their Office of General Counsel, and held some
discussions with OMB staff about the methodology. The information and
documentation we obtained enabled us to generally understand and describe
the methodology and the underlying data, but did not explain all aspects
of the methodology or the specific reasons for certain results. We note in
the report where our description of certain aspects of the methodology is
incomplete. However, these areas were not material to our conclusions.

The primary documentation we initially reviewed, an OMB paper entitled
"Proposal for modification of the ICRAS system," describes (1) OMB's
rationale for adopting its new methodology, (2) analysis OMB performed in
developing the methodology, and (3) certain assumptions and equations that
describe a general theoretical model.1 However, because the paper
describes a theoretical model and includes limited information on specific
analyses performed, data used, key assumptions, and results, and because
not all of the elements of the model described in the paper were used by
OMB, we required additional information from OMB. To obtain additional
information from OMB, we were required to submit written questions to an
attorney in OMB's Office of General Counsel for transmission to OMB
technical staff. The attorney also reviewed the responses that the
technical staff prepared before they were provided to GAO. OMB staff
provided a combination of oral and written responses to our initial set of
questions, but because we still lacked important information, we sought
additional clarification from OMB. At OMB's request, we provided OMB staff
with a Statement of Fact that (1) described our understanding of OMB's
expected loss methodology based on information provided to that point and
(2) identified remaining questions. We met again with the OMB attorney and
technical staff, who responded to our questions. We requested an
electronic version of the methodology, which OMB did not agree to provide.
We attempted to further clarify certain issues, but OMB provided limited
responses. OMB representatives provided certain additional technical
information in comments on a draft of this report.

To determine the estimated default probabilities generated by the default
component of OMB's methodology for fiscal years 2004 and 2005, we adjusted
the expected loss rates for each rating and maturity category that

1The paper that OMB provided was marked "Draft: For Discussion Purposes
Only," but OMB representatives told us there were no subsequent versions
and it should be considered a final paper.

Appendix I
Objectives, Scope, and Methodology

OMB provided to ICRAS agencies by dividing each rate by one minus the
recovery rate OMB assumed for each year. We confirmed that process with
OMB. To determine the extent to which the default probabilities estimated
by OMB's default model differed from the rating agency corporate default
rates used as inputs to the model, we statistically compared the model's
outputs for fiscal year 2004 and 2005 with the corporate default rates
used.

We also determined the current methodology's output in terms of expected
loss rates across the ICRAS risk categories. To do so, we obtained
electronic copies of Ex-Im Bank's cash flow spreadsheets for guarantees as
well as copies of the OMB Credit Subsidy Calculator, which converts agency
cash-flow payments into present value terms. To isolate the default or
expected loss component of Ex-Im Bank subsidy costs from other components
(including fees and interest rate subsidies), we entered consistent
information into the Ex-Im Bank cash flow worksheets for each ICRAS risk
category for fiscal years 2002 through 2005 and conducted this analysis
for 5-year, 8-year, and 10-year credits. We conducted similar analysis for
8-year credits for fiscal years 1997 through 2002. We determined the
present value of the results, based on a constant discount rate, using
OMB's credit subsidy calculator. We discussed our analysis with Ex-Im Bank
officials, who generally confirmed our approach and output.

To determine the impacts of the current OMB methodology on Ex-Im Bank, we
examined changes in the components of Ex-Im Bank's subsidy costs and
financial statement loss allowances. Specifically, we examined the
following:

o 	To determine the effect of the current methodology on Ex-Im Bank's
budget needs, we reviewed Ex-Im Bank budget information from fiscal years
1992-2005 and analyzed changes in the bank's requests for subsidy cost
authorization and its obligation of budget authority for subsidy costs
over that time period. We also interviewed Ex-Im Bank officials regarding
their views on how changes in expected loss rates affected the bank's
subsidy costs. We determined that Ex-Im Bank's budget information was
sufficiently reliable for the purpose of documenting the bank's changing
budget needs and confirmed our analyses with Ex-Im Bank officials. We also
examined Ex-Im Bank information, contained in internal documents, on its
reestimate calculations for fiscal years 19921995 and 1999-2003 and
interviewed Ex-Im Bank officials regarding their views on how changes in
expected loss rates affected the bank's reestimates. Ex-Im Bank's internal
reestimate information differed slightly in some years from information
contained in the Federal Credit

Appendix I
Objectives, Scope, and Methodology

Supplement about the bank's reestimates. Both sources of information
portrayed similar overall trends, but the Ex-Im Bank internal information
covered a longer period of time than the credit supplement information. We
also discussed with Ex-Im Bank officials the bank's process for
calculating reestimates, and we examined auditor workpapers for the 2002
audit of Ex-Im Bank's financial statement, in which the auditors examined
and verified the bank's reestimate for fiscal year 2002. Thus, we
determined that the Ex-Im Bank information was sufficiently reliable for
the purpose of showing trends in the bank's reestimates since the start of
credit reform, as well as the magnitude of the bank's reestimates
following the implementation of OMB's current methodology.

o 	To determine the impact of the current methodology on the changing
relationship between Ex-Im Bank's projection of expected losses and its
fee income, we compared our analysis of expected loss rates for fiscal
years 2002-2005 with the minimum fees that Ex-Im Bank can charge under an
agreement among participating Organization for Economic Cooperation and
Development's (OECD) member countries. We determined these fees using
Ex-Im Bank's Exposure Fee Calculator, available on its Internet site.
Ex-Im Bank officials confirmed our analysis. To identify how the
relationships between expected losses and fee income could be different
for corporate borrowers, we identified the way that corporate ratings are
assigned and fees determined, based on interviews with Ex-Im Bank
officials and on fee information from Ex-Im Bank's Internet site.

o 	To determine the impact of the current methodology on Ex-Im Bank's
financial statement loss allowances, we reviewed Ex-Im Bank's audited
financial statements for fiscal years 2002 and 2003 and the auditor's
workpapers supporting its audit of the 2002 financial statement. We
determined that the data in the audited financial statements were reliable
for the purposes of our analysis. We discussed the modification in Ex-Im
Bank's methodology for calculating financial statement loss allowances,
including the impact of the current methodology's lower loss rates in
calculating those loss allowances, with officials from Ex-Im Bank and its
auditor, Deloitte Touche.

To assess the current methodology and the process by which it was
developed, we identified and evaluated the basis for key OMB assumptions,
methodological components, and data used. We also examined OMB
documentation of the process and discussed the process with

Appendix I
Objectives, Scope, and Methodology

representatives from OMB, Ex-Im Bank, Treasury, and certain other agencies
that participate in the ICRAS process. We did not replicate or validate
the methodology because we lacked complete documentation and did not have
access to the computer programs that were used to estimate OMB's default
model. We also did not determine the reasonableness of specific loss rates
that OMB has estimated.

To assess the methodology, we interviewed cognizant U.S. and foreign
officials and experts and reviewed relevant studies. For example, we
discussed loss estimation methodologies with credit experts and officials
from certain financial institutions, including commercial banks, foreign
export credit agencies, and other foreign officials. On the basis of these
discussions and this review, we identified challenges, concerns, and
practices, such as potential limitations in using certain data for
projecting future defaults and the degree to which institutions followed
similar or different practices in estimating default and loss.

To determine whether the corporate bond default rates used in OMB's
default model have varied significantly since the model was created, we
compared the specific Moody's Investors Service corporate bond default
rates used in OMB's analysis with updated published versions of those
Moody's rates.

We obtained information from OMB regarding its comparison of Ex-Im Bank
default rates to the corporate default data used in its model, and we
obtained from Ex-Im Bank the historical data sets that Ex-Im Bank said it
gave OMB. We obtained certain information from OMB about how it analyzed
the Ex-Im Bank data, and while we were able to determine the time period
primarily covered by this analysis, we were not able to determine the
specific data that OMB analyzed because we were unable to reconcile
certain information that OMB provided. In assessing the time period
covered by OMB's analysis, we obtained historical country ratings data
from Moody's and Standard & Poor's documents and historical ICRAS ratings
information from Ex-Im Bank. We used these to determine the proportion of
observations in the Ex-Im Bank data sets for which risk ratings at the
time of the transaction were available.

We examined OMB's assumptions about recovery rates and compared these with
aggregate recovery rates that we calculated on Ex-Im Bank datasets
covering guarantees and some insurance for 1985-1999 and 19852001, as well
as with recovery rate assumptions made by other financial institutions. We
discussed the reliability of the Ex-Im Bank data sets and of

Appendix I
Objectives, Scope, and Methodology

the underlying systems used to create the data with Ex-Im Bank officials,
who said they view the data they have compiled to be a reasonable
representation of their historical experiences and adequate for its
intended purposes, which were initially to provide information about Ex-Im
Bank's activities to a potential private sector partner. The officials
stated that the reliability of data about individual transactions is
considerably greater for transactions initiated in 1996 and later because
of changes that improved data entry and verification. We determined the
data were sufficiently reliable for our purpose of comparing aggregate
recovery rate information in the datasets with the recovery rate
assumptions used by OMB.

To broadly assess the technical features of OMB's default model, we
evaluated information provided by OMB that described the model's equations
and how they were estimated, based on standard econometric criteria. We
did not conduct a complete technical review because we did not have access
to full documentation of the model or the model in electronic format.

To assess the process by which the current methodology was developed, we
discussed with OMB representatives and certain other ICRAS officials the
respective agencies' role and degree of involvement in developing and
providing comment on the methodology. We also reviewed documents that OMB
had distributed to ICRAS agencies about its methodology and discussed with
ICRAS officials the general time frames in which the methodology was
developed and the nature of certain meetings that OMB held to present
information about its methodology. The ICRAS officials we interviewed
received information about the methodology's development and
implementation and have had continuing participation in the ICRAS process.
We also reviewed credit reform guidance on preparing and auditing subsidy
costs.2

2Federal Accounting Standards Advisory Board, Federal Financial Accounting
and Auditing Technical Release 6-Preparing Estimates for Direct Loan and
Loan Guarantee Subsidies under the Federal Credit Reform Act, Amendments
to Technical Release 3: Preparing and Auditing Direct Loan and Loan
Guarantee Subsidies Under the Federal Credit Reform Act (Washington, D.C.:
January 2004). This guidance was developed by the Accounting and Auditing
Policy Committee, a permanent committee of the Federal Accounting
Standards Advisory Board. The committee was organized by OMB, GAO, the
Department of the Treasury, the Chief Financial Officers' Council, and the
President's Council on Integrity and Efficiency, to research accounting
and auditing issues requiring guidance.

Appendix I
Objectives, Scope, and Methodology

To provide information on the reserve practices of foreign ECAs, we
judgmentally selected a sample of four ECAs that are key competitors of
Ex-Im Bank or that were identified by knowledgeable U.S. and private
sector officials as entities that had examined or changed their reserve
practices in recent years. These included Compagnie Franc,aise d'Assurance
pour le Commerce Exterieur in France, Euler Hermes
Kreditversicherungs-Aktiengesellschaft in Germany, the Export Credits
Guarantee Department in the United Kingdom, and Export Development Canada.
In each case, we discussed with officials, and reviewed available
documentation on, the ECA's statutory mandate, financial activities, and
reserve practices. We also reviewed public financial statements where
available. We also met with officials from other government organizations
in these countries, including treasury or finance ministries. We met with
officials from the Office of the Auditor General of Canada, France's Cours
des Comptes, and the U.K.'s National Audit Office, because those offices
audit the financial statements of the Canadian, French, and U.K. ECAs,
respectively. We obtained the perspectives of officials from ECAs and
other government agencies on the difficulties associated with developing
loss estimation methodologies and using available data. In addition, we
discussed these issues with an official from the export credit group of
the OECD, and officials at the Office National du Ducroire/Nationale
Delcrededienst in Belgium, including the chair of a group of OECD export
credit country risk experts.

To provide information on the reserve, or loan loss allowance, practices
of commercial banks, we judgmentally selected a sample of three U.S.
commercial banks with large lending portfolios totaling approximately $800
billion, including large international exposures. For each bank, we spoke
with management involved in international lending and the calculation of
the bank's loan loss allowance. In addition, we reviewed the banks'
financial statements and any documentation that was provided. We also met
with several U.S. banking regulators-the Federal Reserve Board, Office of
the Comptroller of the Currency, and the Federal Deposit Insurance
Corporation-to discuss the loan loss allowance guidance banks are required
to follow. We reviewed both regulatory and accounting guidance governing
the calculation of the loan loss allowance by commercial banks.

Appendix II

Loss Estimation Practices of Foreign Export Credit Agencies

Significant variation exists among the loss estimation, or reserve,
practices of foreign export credit agencies (ECA) that we consulted.1
Differences in mission, structure, and accounting approaches help explain
this variation. Some of these ECAs are expected to avoid competing with
private sector financial institutions, which may result in more exposure
to emerging market borrowers and riskier portfolios as compared with other
ECAs. These ECAs' financial relationships with their governments differed,
as did their individual responsibility for covering any losses that might
result from their activities. Some ECAs follow an accounting approach that
prescribes estimating probable losses over time, while others follow an
accounting approach that precludes such estimation. ECAs in Canada and the
United Kingdom (U.K.) have recently adopted or plan to implement new
methodologies for estimating the likelihood of default and loss associated
with their activities. ECAs in France and Germany follow a simpler
approach in which the fees they collect on a given transaction, in
accordance with an international agreement among export credit agencies,
are regarded as sufficient to cover any likely losses on the transaction.
The French ECA is studying a new accounting system that would enable it to
more closely align its loss expectations with its historical repayment
experiences. (App. I contains information about our objectives, scope, and
methodology for examining the reserve practices of foreign ECAs.)

ECAs in Canada and the United Kingdom Have Methodologies to Estimate
Future Defaults and Losses in Determining Reserve Levels

ECAs in Canada and the United Kingdom determine their required level of
loss reserves by estimating the extent of probable losses in their
portfolio at a point in time, some of which may result from future
defaults. These ECAs' missions, structures, and accounting approaches lend
important context to their reserve practices. The Canadian and U.K. ECAs
have recently revised (or are in the process of revising) their risk
assessment and reserve practices to more precisely measure future losses.
In developing their approaches, these two ECAs examined the risk
assessment and reserve practices of leading financial institutions and
worked with private sector risk assessment specialists. Their approaches

1We discussed reserve practices with Export Development Canada, Compagnie
Franc,aise d'Assurance pour le Commerce Exterieur in France, Euler Hermes
Kreditversicherungs-AG in Germany, the Export Credits Guarantee Department
in the United Kingdom, and the Office National du Ducroire/Nationale
Delcrederedienst in Belgium. For ease of reference, we do not use these
entities' proper names in this appendix. We recognize that "reserves" is
not a technical term, but we use it in this appendix because each ECA used
different terminology to refer to its practices.

                                  Appendix II
                  Loss Estimation Practices of Foreign Export
                                Credit Agencies

have similar elements but differ in important respects. The new approaches
have been reviewed by specialists outside the ECA.

Mission, Structure, and Accounting Approaches Affect Reserve Practices

The Canadian and U.K. ECAs' mission is to help facilitate national
exports, but their particular methods and structure for doing so differ.
The Canadian ECA is a wholly owned government corporation that was
capitalized with funds from the Canadian government and operates with the
full faith and credit of the Canadian government. According to officials
of this ECA, the entity is self-sustaining, in that it does not receive
annual infusions of budgetary support for its operations or losses. Its
largest business activity in terms of volume is short-term export
insurance, but it also offers loans and medium-term insurance and loan
guarantees. According to the Canadian ECA, it takes a commercial approach
to managing its risks to ensure its long-term financial health. This
institution makes its own decisions about the credits it will offer and is
not prohibited from competing with private sector financial institutions.2
Long-term transactions that it determines are beyond its risk capacity and
are inconsistent with its long-term health may be referred to the
government of Canada for consideration. The Canadian government may accept
and manage those risks provided that there is sufficient national benefit
to Canada.3

In contrast, the U.K. ECA is a government department that receives annual
budgetary support (subject to Parliamentary approval) to help fund its
operations and cover its losses. The U.K. ECA is in a time of transition.
The ECA's operations were streamlined in 1991, when new legislative
authority required it to sell its short-term insurance business and focus
on mediumand long-term project finance.4 The U.K. ECA is expected to avoid
direct competition with U.K. private insurers and banks. It is also
expected to undertake and manage its activities to ensure, with a high
degree of

2The majority of business the Canadian ECA undertakes annually is
transacted in what it calls its Corporate Account.

3These are known as Canada Account transactions and are undertaken
infrequently. The Canadian ECA executes the transaction on behalf of the
government.

4Export and Investment Guarantees Act 1991.

Appendix II
Loss Estimation Practices of Foreign Export
Credit Agencies

confidence, that it will break even financially.5 Simultaneous attainment
of these two objectives suggests this ECA has to strike a balance in the
types of transactions and the nature of risks it shall undertake. The
ECA's risk premia and larger transactions are subject to approval by the
U.K. treasury department. The level of treasury department control
increased following certain losses the ECA incurred in the late 1990s.
Plans are under way to convert the U.K. ECA into a separately capitalized,
self-sustaining entity that would operate at arm's length from the U.K.
government, responsible for managing its own financial losses.6

Different ECA missions and structures can have implications for ECA risk
profiles and, thereby, reserve practices. For example, with its broader
mandate, about 60 percent of the Canadian ECA's 2003 portfolio exposure
was to U.S. and Canadian borrowers. In contrast, most of the U.K. ECA's
ten most active markets in fiscal years 2001 and 2002 were emerging
markets, representing about three-fourths of its activity in these years.7
These different risk profiles affect the level of loss reserves that these
ECAs hold. According to the Canadian ECA's financial statements, its
allowances for loss averaged about 10 percent of its total exposure during
calendar years 2000-2003. According to the U.K. ECA, its allowances for
loss averaged about 20 percent of total exposure between fiscal years
20002003.8 It is important to note that, for business undertaken since
1991, the U.K. ECA is expected to maintain reserves equal to at least 150
percent of

5According to U.K. officials, a mandate to break even over the long term
has been in effect since the ECA's inception, but over time this has been
translated into quantifiable objectives. This mandate currently applies to
all business undertaken since 1991 that meets the ECA's underwriting
criteria. The mandate does not apply to business that the ECA is directed
to undertake on the basis of national interests but that does not meet its
underwriting criteria.

6According to U.K. ECA officials, the conversion of their entity to a
self-sustaining "Trading Fund" is scheduled to take place by 2007.

7The U.K. ECA's fiscal year begins on April 1 and ends on March 31. The
above figure represents the ECA's activity for the fiscal years ending
March 31, 2002 and 2003.

8This figure reflect allowances on the U.K. ECA's pre-1991 and post-1991
activities and includes allowances on paid claims and future amounts at
risk.

Appendix II
Loss Estimation Practices of Foreign Export
Credit Agencies

expected losses.9 In comparison, the U.S. Export-Import Bank's (Ex-Im
Bank) loss allowances have averaged about 18 percent of its total exposure
since fiscal year 1999.

Moreover, different degrees of government support affect the extent to
which ECAs are responsible for managing their own financial risk and
protecting taxpayers from loss. For example, as a self-sustaining entity,
the Canadian ECA does not receive annual budgetary support from its
government and would be expected to cover any losses it incurs.10 In
contrast, although the U.K. ECA is expected to break even over time, it
receives appropriations from the U.K. Parliament to cover anticipated
losses and administrative expenses. It also operates with a treasury
department guarantee of the obligations arising from its guarantees.

Both the Canadian and the U.K. ECAs follow accrual-based accounting
standards, in which revenue and expenses are recorded in the period they
are earned or incurred, even though they may not have been received or
paid.11 Under such accounting, loss reserves are an estimate of probable
losses in a portfolio as a whole. The reserves are normally recorded long
before actual defaults occur. This contrasts with cash flow accounting, in
which revenue is recognized when it is received and expenses when they are
paid. As discussed in the section on the French and German ECAs, this
method of accounting precludes the matching of revenue and expenses over
time.

9The U.K. ECA must achieve a "Reserve Coverage Ratio" of at least 1.5
times the level of future expected losses in its portfolio. Because the
U.K. ECA is expected to generally break even, including during periods
when losses are concentrated and thus unusually high, it is expected to
reserve at a level that is higher than would be needed to cover expected
losses on average over a long period. The extra reserves are to cover what
are sometimes called "unexpected losses." With these higher reserve
levels, reserves should be adequate to cover losses 75 percent of the
time, according to U.K. ECA officials.

10According to the Canadian ECA, its overall operations have resulted in a
profit in every year but one.

11The Canadian ECA is required to follow Canadian Generally Accepted
Accounting Principles (GAAP), which are similar to U.S. private-sector
GAAP and include the use of accrual-based accounting. The U.K. ECA follows
U.K. government accounting standards, which are accrual-based standards.
This ECA practiced cash flow accounting for many years but switched to
accrual-based accounting several years ago.

                                  Appendix II
                  Loss Estimation Practices of Foreign Export
                                Credit Agencies

Canadian and U.K. ECAs Have Recently Adopted, or Will Implement, New
Reserve Methodologies

Determining Risk Ratings and Calculating Probability of Default

The Canadian and U.K. ECAs have recently revised their existing reserve
practices by adopting or moving toward implementing methodologies that are
designed to more precisely measure risk in their portfolios and ensure
that their reserves reflect those risks. In 2001, the Canadian ECA and the
Canadian Office of the Auditor General, who audits the ECA's financial
statements, undertook a review of the ECA's reserve methodology with the
goal of making it reflect current best practices. The ECA studied the
reserve practices of several leading U.S. and Canadian banks and other
ECAs, including Ex-Im Bank, and examined new developments in bank
regulatory and accounting guidance.12 According to Canadian ECA officials,
it adopted a new risk assessment model that follows the risk assessment
approaches used by some other financial institutions with international
risk exposures. The new methodology did not have a substantial impact on
the ECA's level of reserves, although the entity changed the process by
which it calculated its reserves, specifically its components and what it
covers. For example, it began establishing reserves for committed
undisbursed credits, which it had not done previously.

In 1999 the U.K. treasury department hired a private consulting firm with
credit risk expertise to review the effectiveness of the ECA's risk
management systems because of concerns about the ECA's financial
condition, given the larger than expected losses it incurred during the
Asian financial crisis. According to U.K. ECA officials, the consulting
firm concluded that the ECA's process for estimating expected loss was
reasonable but recommended, among other things, that the U.K. ECA should
better assess the risk of, and establish capital to buffer against,
unexpected losses.13 The U.K. ECA is upgrading its existing risk
assessment models and processes in response to the review.

The Canadian and U.K. ECAs each use a combination of rating agency data
and their own analyses and adjustments in determining ratings levels and
default probabilities. For rating corporate risk, the Canadian ECA uses
ratings from major rating agencies such as Moody's Investors Service and

12The Canadian ECA examined the Ex-Im Bank loss estimation methodology in
place at the time of its benchmarking exercise that began in 1999, which
differed from the loss estimation methodology implemented by the Office of
Management and Budget for fiscal year 2003 that is described in this
report.

13Risk Management Review for HM Treasury and ECGD, KPMG, December 1999.

Appendix II
Loss Estimation Practices of Foreign Export
Credit Agencies

Standard & Poor's when they are available; when these ratings are not
available, the Canadian ECA's risk department assigns ratings using
standard rating agency criteria. They place credits into seven risk
categories. The Canadian ECA then estimates its default probabilities for
its corporate borrowers using published default rates from Moody's
Investors Service and Standard & Poor's, taking into account the maturity
of the credits. The Canadian ECA rates sovereign borrowers based on its
own research and country knowledge. Once these ratings are assigned, the
Canadian ECA uses the default probabilities associated with those ratings
from Standard & Poor's and Moody's in determining default probabilities.
For both corporate and sovereign borrowers, the Canadian ECA adjusts
rating agency default probabilities where it believes such adjustments are
necessary.

For determining sovereign default probabilities and risk ratings, the U.K.
ECA uses a model it developed in 1991 that assesses countries' likelihood
of default, using macroeconomic data such as borrower country
indebtedness. Its analysts consider the model's output and additional
factors, including other country data, rating agency sovereign ratings and
interest rate spreads, in the final assignment of sovereign ratings. For
determining expected loss, the expected duration of default periods and
the recovery rates when defaults occur are taken into account, along with
the probability of default.

For rating corporate transactions, the U.K. ECA uses rating agency
corporate risk ratings where they are available. For corporations that are
not rated by the major rating agencies, the U.K. ECA assigns ratings using
templates developed with a major rating agency. In both cases, once these
ratings are obtained, U.K. ECA officials adjust them, in some cases, based
on a comparison with country sovereign ratings. For assigning expected
losses to different risk ratings, U.K. ECA officials use a Standard &
Poor's tool that is based on that rating agency's historical data on
ratings transition and default rates and that incorporates other
information such as recovery rates.

U.K. ECA officials are moving toward a new modeling process that will
directly assess, not only the risk of expected loss, but also the capital
needed to cover unexpected losses or the risk of having greater losses
concentrated in certain periods. This model was developed in consultation
with a U.K. credit risk expert and uses a combination of private ratings
agency data on sovereign bond defaults from the 1990s and U.K. ECA data on
the 1980s default experience of the U.K. ECA.

                                  Appendix II
                  Loss Estimation Practices of Foreign Export
                                Credit Agencies

Calculating Expected Loss	Once default probabilities have been determined,
determining the expected losses from the defaults involves determining the
likely amount of loss when defaults occur, based on expected recoveries.
The Canadian ECA uses its own historical data, where sufficient, to
estimate recoveries for sovereign and nonsovereign borrowers. The U.K. ECA
determines its own recovery rates for sovereign borrowers; for corporate
borrowers, it makes some use of rating agency recovery data.14 Both ECAs
assume higher losses (lower repayments) for corporate than sovereign
defaults.15 The Canadian ECA also assumes that higher losses will be
incurred from defaults on unsecured credits than on collateralized
credits. The calculation of expected loss forms these entities' base
reserves amount, to which certain upward adjustments are made. These
adjustments reflect the potential unexpected losses that are affected by
the portfolio effects of concentration and correlation of financing
activities.

Adjusting for Portfolio Risk	The Canadian ECA follows a portfolio approach
in estimating loss and calculating reserves. In such an approach, the base
reserve amount is adjusted upward because of additional risks related to
concentrations of exposure and correlations among credits. The Canadian
ECA adds reserve amounts for significant exposures to single borrowers,
countries, or industries. It also adjusts upward for the possibility that
problems in one area (for example, in one country) will spread to other
parts of its portfolio.

The U.K. ECA's current approach includes some judgmentally determined
upward adjustments to its base expected loss calculations for certain
concentrations of risk. These include upward adjustments for public,
nonsovereign borrowers, as well as certain systemic risks related to other
countries or industries. Its new risk management approach will make such
adjustments more systematically, as part of its loss model.

Determining Fees	Canadian and U.K. ECA officials both said that their loss
estimation methodologies help them determine what fees to charge borrowers
for their products. Both ECAs agree to follow a risk rating and pricing
agreement developed by participating Organization for Economic Cooperation
and Development (OECD) countries, in which participants

14For aircraft financing, the U.K. ECA uses data on the recovery or
second-hand value of aircraft.

15For the Canadian ECA, secured commercial debt has a lower probability of
loss given default than does sovereign debt.

                                  Appendix II
                  Loss Estimation Practices of Foreign Export
                                Credit Agencies

jointly develop country risk ratings for the purposes of determining the
minimum fees to be charged at each risk rating.16 (More information about
this agreement is provided below.) However, according to Canadian and U.K.
ECA officials, they may apply higher fees if they determine that the fees
do not adequately reflect their own assessment of the potential loss on a
transaction. Officials from both ECAs stated that setting fee levels in
relation to expected loss is an important component of their financial
stability over time.

Canadian and U.K. Reserve Methodologies Were Subject to Internal and
External Review

According to officials from the Canadian and U.K. ECAs, their reserve
methodologies (including key assumptions and computer models) have been or
are being reviewed extensively, both internally and externally, before
being adopted. The Canadian ECA's new methodology was positively reviewed
by an independent national accounting firm with significant experience in
reviewing loan loss methodologies. The U.K. ECA's current reserve
practices were subject to review by a private consulting firm. In
responding to the consulting firm's recommendations, the ECA has worked
with the U.K. treasury, a prominent academic and credit risk expert, and a
private rating agency to develop its new approach. Its new risk assessment
model is based on a well-known credit risk model that was developed by a
leading U.S. bank.17

ECAs in France and Germany Use Fees to Offset Loss

ECAs in France and Germany base their reserve practices primarily on the
fees they collect for their products rather than on systematic estimations
of their probable losses. These ECAs are private companies that offer
export credit insurance on behalf of their respective governments,
following certain risk thresholds that their governments establish for
these accounts. Their governments expect these accounts to break even in
the long run. The French and German ECAs are not expected to estimate
future losses when transacting new business but rely instead on the OECD
participating countries' guidance in setting their fees. Neither France
nor Germany annually appropriates budgetary funds to cover loss; instead,
the fees the

16This group, the Participants to the Arrangement on Officially Supported
Export Credits, is not an official OECD body, but receives support from
the OECD Secretariat. Its members include Australia, Canada, the European
Community, Japan, Korea, New Zealand, Norway, Sweden, and the United
States.

17The underlying model, CreditMetrics, was introduced by JP Morgan in
1997.

                                  Appendix II
                  Loss Estimation Practices of Foreign Export
                                Credit Agencies

ECAs collect constitute these countries' reserve against future loss. The
government ministries that oversee the ECAs conduct independent
assessments of country risk in setting or adjusting the risk thresholds
that specify the degree to which they will offer credit in certain
countries. Further, the French ECA is developing a method, not yet in
place, to more independently estimate future losses on the government's
export credit activities and to track the actual losses incurred over
time.

French and German ECAs Provide Export Insurance and Guarantees on Behalf
of Their Governments and Follow Government Accounting Methods

France and Germany provide and account for their official export credit
business in similar ways. In both countries, the official ECA is a private
enterprise that insures or guarantees exports on behalf, and at the
direction, of the government. In addition to managing their government's
export credit business (referred to in both cases as the state account),
the French and German ECAs also engage in business for their own account.
The French and German state accounts extend primarily medium-and long-term
export credit insurance, whereas the private enterprises that manage the
state accounts primarily sell short-term insurance.18 The ECAs are not
responsible for any profit or loss incurred on the state account, which
transfers to the government.19 The French and German ECAs both receive
administrative fees from the government for their services.

The government ministries that oversee the ECAs make decisions about the
degree to which the state accounts will offer export credits in certain
countries or to certain borrowers (exposure limits are discussed further
below). French and German state accounts are expected to operate in
riskier markets where private export credit insurance cannot be obtained.
In both countries, the government ministries can also direct the ECA to
undertake certain transactions, even if doing so will cause it to exceed
established exposure limits, if the government believes the transactions
to be in the country's best interest.

The French and German state accounts are both directed to break even over
the long term, meaning they should incur neither large gains nor losses.
However, assessing compliance with this mandate is difficult,

18The French and German state accounts do not directly finance exports.
Private financial institutions handle this instead.

19According to French and German ECA and government officials, the state
accounts are operated separately from ECAs' private account. Profit on the
private account is not used to fund the state account.

                                  Appendix II
                  Loss Estimation Practices of Foreign Export
                                Credit Agencies

because both governments' budgets are accounted for on a cash flow basis.
Under cash flow accounting, revenue is recognized when it is received, and
expenses are recognized when they are paid. Thus, revenue in a given year
is compared with expenses in that same year, regardless of when the
underlying transactions occurred. Under this system, it can be difficult
to know the degree to which the fees collected in a given year from
providing export credit insurance cover any claims made against that
insurance in later years. French and German ECA officials acknowledged
that this accounting practice limits their ability to fully analyze actual
or potential losses on the state account. The French ECA has been
developing an alternative accounting approach that would enable such
analysis, as discussed later.

French and German ECAs Follow Government-Determined Risk Thresholds and
Use Fees to Cover Loss

The French and German approaches to evaluating risk are based primarily on
assessing country risk for the purpose of setting exposure limits and
using fees to cover losses. The French and German ECAs are not expected to
estimate expected losses in advance of undertaking transactions, but the
French ECA provides informal risk assessments to the oversight ministry
for it to consider when making decisions about undertaking transactions.
The French and German governments do not provide budgetary funds at the
beginning of a year to cover any losses that might be incurred during the
year, but any losses incurred on the state account are automatically
covered.

The government ministries that oversee the French and German ECAs analyze
borrower countries' risk profiles when making their annual decisions about
the exposure limits that will be in effect for a given year. This analysis
may consider macroeconomic factors, OECD country risk ratings, and the
ECA's experiences with other countries' repayment histories on previously
extended credit. In France, the ECA also provides input to this analysis.
These exposure limits, or ceilings, are developed to ensure portfolio
diversification and constrain the accumulation of excessive risk levels.
In both countries, the government determines risk ceilings and provides
these to their ECAs to follow in operating the state account. Country risk
ceilings can be exceeded only at the government's direction. The German
ECA also faces a statutory limit on the total exposure it can undertake in
a given year.20

20A German ECA official told us that the statutory limit has never
precluded it from undertaking any business that it wanted to undertake for
the state account.

Appendix II
Loss Estimation Practices of Foreign Export
Credit Agencies

In both countries, the fees collected on the export credit business are
viewed as a reasonable approximation of the amount of loss likely to be
incurred. In setting fees, both ECAs follow the guidance agreed to by the
participating OECD countries for assigning risk ratings to sovereign
borrowers and charging premium rates. These ECAs add surcharges to the
OECD minimum fee rates for corporate borrowers, recognizing the higher
risk of corporate business. As discussed below, the OECD fees were
politically determined, and both entities have considered whether the
premia are sufficient to cover actual losses.

France and Germany differ in how they manage their fee revenue. In both
countries, the revenue belongs to the government. However, in France, some
fee revenue is kept with the French ECA for the purpose of paying expenses
that are incurred on the state account, such as paying a claim on a
defaulted credit. Officials of the French ministry that oversees the ECA
told us that the amount left on deposit with the ECA is based on their
best estimate of the losses that the ministry expects will materialize in
a given year. A German ECA official said that all fee revenue it collects
is immediately transferred to the German government. When the German ECA
has to pay a claim, it must request funds from the German government.

French ECA and government officials told us that the French ECA is
developing a new accounting system for the state account that will enable
it to analyze, for each underwriting year, the collected fees and the
claims corresponding to the transactions covered during the same year.
This system will provide the ECA with historical data on payment
experience in order to calculate expected losses on a statistical basis.
However, the process is not complete, and the information it has produced
is not used for official purposes. The approach in development still uses
OECD minimum fees as the baseline for estimating loss but will add
surcharges to take account of increased risk, for example, when a default
is pending. Surcharges will also be added for risks other than sovereign
risks. According to French ECA and government officials, a key challenge
in developing this approach was compiling payment histories for individual
transactions. They also stated that before the French state account would
develop a reserve system that does not rely on the OECD minimum fees,
further accumulation of historical data on payment experiences will be
necessary, a process they expect will take some time.

                                  Appendix II
                  Loss Estimation Practices of Foreign Export
                                Credit Agencies

OECD Participating Countries' Risk Assessment and Fee Arrangement

The agreement of the participating OECD countries regarding country risk
classifies countries into seven risk categories on the basis of a country
risk assessment model. The model ranks countries according to which is
most likely to default, considering indicators of countries' financial and
economic situations. It also uses data provided since 1999 by export
credit agencies on their payment experiences with countries. Quantitative
scores for each country determine its initial risk classification, which
can be adjusted by participating countries based on an assessment of
political risk and other factors not considered in the model itself.

While the model scores provide some indicator of default probabilities for
each country, they are not used in determining what the risk premiums, or
fees, should be to cover expected loss for financing to sovereign buyers
in a risk category. The fees result from a political agreement, an
averaging of fees in place in 1996 across member export credit agencies.
Thus, the fees reflect the expected loss of lending to sovereign borrowers
at various risk levels only to the extent that the average of 1996 fees
across participating countries reflect those expected losses. The minimum
common fees are not intended to reflect the generally higher risk of
lending to nonsovereign, or private, borrowers within the same country,
according to OECD and ECA officials.

Participating OECD countries are collecting data from their ECAs on their
financing and repayment experiences since 1999, in order to assess the
validity of the current fees as indicators of expected loss. According to
an ECA official who chairs a group of country risk expert from OECD
countries, collecting enough data to assess the current premiums will take
at least 10 years.

Appendix III

Loan Loss Allowance Guidance and Select Commercial Bank Practices

Loans are the largest component of most depository institutions' assets;
therefore, the loan loss allowance1 is critical to understanding the
financial condition of a depository institution and changes in credit
risks and exposures. Given the importance of the loan loss allowance as an
indicator of financial condition, and because adjustments to the allowance
affect an institution's earnings, the loan loss allowance is scrutinized
by regulatory agencies. Regulators and the accounting profession
acknowledge that calculating the loan loss allowance requires significant
judgment and that accounting and regulatory guidance are not prescriptive.
For the past several years, the organizations involved in developing U.S.
private sector accounting standards-the Financial Accounting Standards
Board (FASB) and the American Institute of Certified Public Accountants
-have been in the process of reviewing current loan loss allowance
guidance. Likewise, the U.S. federal banking regulators-the Federal
Reserve Board (FRB), Office of the Comptroller of the Currency (OCC),
Federal Deposit Insurance Corporation (FDIC), Office of Thrift
Supervision, and the National Credit Union Administration-and the
Securities and Exchange Commission have been updating regulatory guidance
governing the loan loss allowance. While the commercial banks we contacted
follow the basic concepts of accounting and regulatory guidance, specific
aspects of these banks' allowance methodologies differ. Regulatory
guidance requires U.S. banks involved in international lending to address
additional risks, in addition to the accounting and loan loss allowance
concepts that apply to domestic lending.

Loan Loss Allowance Is an Important Factor in an Institution's Financial
Condition

The loan loss allowance plays a key role in the financial condition of a
bank.2 It reflects a bank's judgment of the overall collectibility of its
loan portfolio; that is, the higher the percentage of a bank's loan loss
allowance to its total loan portfolio, the lower the estimated
collectibility of the loan portfolio and the higher the estimated level of
credit risk.3 It also reflects

1Different phrases are used interchangeably when discussing credit loss
reserves, including loan loss reserves, allowance for loan and lease
losses, and allowance for credit losses. We will use the phrase "loan loss
allowance."

2Loan loss allowance guidance applies to banks and other financial
institutions. We discuss banks in this appendix because our review focuses
on lending done by commercial banks for purposes of comparison with the
Export-Import Bank.

3Credit risk is the potential for financial loss resulting from the
failure of the borrower or counterparty to perform on an obligation.

 Appendix III Loan Loss Allowance Guidance and Select Commercial Bank Practices

the amount of estimated losses that have occurred in the loan portfolio
but have not yet been realized. The loan loss allowance, according to bank
regulatory guidance, must be appropriate to absorb estimated credit losses
inherent in the loan portfolio. When changes are made in the loan loss
allowance, these changes directly affect an institution's earnings. The
loan loss allowance is established and maintained by charges against the
bank's operating income, which reduces earnings,4 or by reversals of the
allowance that would increase earnings.

Given the loan loss allowance's effect on earnings and the role the
allowance plays in allowing banks to cover probable and estimable losses,
U.S. financial regulatory agencies pay close attention to a bank's loan
loss allowance. These regulators require banks to establish and regularly
review the adequacy of their allowance, and bank examiners assess the
asset quality of an institution's loan portfolio and the adequacy of the
loan loss allowance. Because regulatory guidance is not prescriptive, bank
regulators told us that, through examinations, they assess the
"reasonableness" of a bank's loan loss allowance by comparing a bank's
loan loss allowance level with industry standards and looking for
justification for any methodology that could be considered an outlier. As
part of their assessment of public company filings, the Securities and
Exchange Commission also reviews banks' loan loss allowance disclosures.

Accounting and Regulatory Guidance Are Not Prescriptive; the Loan Loss
Allowance Requires Significant Judgment

Regulatory agencies told us that their guidance is not prescriptive, and
accounting and regulatory guidance states that the loan loss allowance
requires a significant amount of judgment. Because no single approach has
been determined to encompass the wide variety of banks' loan portfolios
and their varying degree of risk and unique historical loss experience,
regulatory and accounting guidance provide principles and guidelines for
banks to follow, rather than specific formulas and factors for banks to
use in their allowance calculations. The bank regulators direct
institutions to follow U.S. generally accepted accounting principles
(GAAP), as it applies to the loan loss allowance, for regulatory reporting
purposes. Specifically,

4The management of a bank adjusts the level of its loan loss allowance
through periodic provisioning to offset charge-offs to reflect the level
of estimated losses in the loan portfolio. Provisions are an expense
charged against an institution's current earnings and represent the amount
necessary to adjust the loan loss allowance to reflect probable and
estimable uncollectible loan balances.

Appendix III Loan Loss Allowance Guidance and Select Commercial Bank
Practices

banks follow Statement of Financial Accounting Standards (SFAS) 114,
Accounting by Creditors for Impairment of a Loan, in estimating losses
from individual impaired5 loans. Further, SFAS 5, Accounting for
Contingencies,6 provides guidance to banks in their calculation of losses
for pools of loans, impaired or performing, which are evaluated
collectively.

The bank regulators we spoke with-FRB, OCC, and FDIC-stated that
regulatory guidance is coordinated across all the banking regulatory
agencies and is consistent with GAAP. On March 1, 2004 the banking
regulators issued an Update on Accounting for Loan and Lease Losses, which
addresses recent developments in accounting for the loan loss allowance
and presents a list of the current sources of GAAP and supervisory
guidance for accounting for the loan loss allowance. One of the sources it
lists is The Interagency Policy Statement on the Allowance for Loan and
Lease Loss, which was issued by FRB, OCC, FDIC, and the Office of Thrift
Supervision in 1993. This document discusses the nature and purpose of the
loan loss allowance, the responsibilities of a bank's board of directors
and management, how banks should determine the adequacy of their allowance
and the factors that should be considered in their estimates, and
examiners' responsibilities with regard to the loan loss allowance. Prior
to the March 2004 Update on Accounting for Loan and Lease Losses, the 1993
interagency policy was supplemented by the 2001 Federal Financial
Institutions Examination Council7 Policy Statement, discussed later.

In addition to the interagency policy, OCC and FDIC issue their own loan
loss allowance policy statements that they distribute to banks under their
supervision. These statements are in line with the interagency policy.

5A loan is impaired when, based on current information and events, it is
probable that a creditor will be unable to collect all amounts due
according to the contractual terms of the loan agreement.

6SFAS 5 defines a contingency as an existing condition, situation, or set
of circumstances involving uncertainty as to possible gain or loss to an
enterprise that will ultimately be resolved when one or more future events
occur or fail to occur.

7The Federal Financial Institutions Examination Council is an interagency
body empowered to prescribe uniform principles, standards, and report
forms for the federal examination of financial institutions by the federal
banking regulators and to make recommendations to promote uniformity in
the supervision of financial institutions.

 Appendix III Loan Loss Allowance Guidance and Select Commercial Bank Practices

Table 1 provides a summary of the relevant accounting and regulatory
guidance governing the loan loss allowance.

  Table 1: Summary of Accounting and Regulatory Guidance Followed by Banks in
                     Their Loan Loss Allowance Calculation

Accounting guidance Regulatory guidance

                           Individual impaired loans

Conditions: Under SFAS 114, a loan is impaired when it is Conditions:
Banks determine impairment using their loan review probable that the bank
will be unable to collect amounts due procedures. Generally, a loan is
impaired for the purposes of according to the terms of the loan. Banks
determine impairment SFAS 114 if it exhibits the same level of weaknesses
and using normal loan review procedures. probability of loss as loans (or
portions of loans) classified as

"doubtful" or "loss." (See table 2 for risk classifications.)

Measurement of impairment: Under SFAS 114, one of the Measurement of
impairment: SFAS 114 guidance should be following methods must be used:
followed. Regulators expect that loan loss allowances for all

o  present value of expected future cash flows discounted at the impaired,
collateral dependent loans will be based on the fair loan's effective
interest rate, value of the collateral for purposes of regulatory reports.

o  loan's observable market price, and

o  fair value of collateral if the loan is collateral dependent.

Pools of loans

Conditions: Under SFAS 5, the following conditions must be met:
Conditions: Banks should follow guidance in SFAS 5 for
(1) it is probable that a loan has been impaired at the financial
measuring losses for pools of loans.
statement date and (2) the amount of loss can be reasonably
estimated. Under SFAS 114, some impaired loans may have risk
characteristics in common with other impaired loans, as such a
bank may aggregate those loans for evaluation of impairment.

Measurement of impairment: According to SFAS 5, whether the amount of loss
can be reasonably estimated will depend on, among other things, the
experience of the bank, information about the ability of individual
debtors to pay, and analysis of the loans in light of the current economic
environment. In the case of a bank with no relevant experience, reference
to the experience of other enterprises in the same business may be
appropriate. According to the interpretation of SFAS 5 (FASB
interpretation 14), when a reasonable estimate of a loss is a range and
when some amount within the range appears at the time to be a better
estimate than any other amount within the range, that amount will be
accrued. When no amount within the range is a better estimate than any
other amount, the minimum amount in the range will be accrued.

Measurement of impairment: Because no single approach has been determined
to be appropriate for all banks, a specific method to determine historical
loss experience is not required.

o  The method a bank uses will depend to a large degree on the
capabilities of its information systems.

o  Acceptable methods range from a simple average of the bank's historical
loss experience over a period of years to more complex "migration"
analysis.

o  There is no fixed historical period that should be analyzed by banks to
determine average historical loss experience.

          Source: GAO analysis of accounting and regulatory guidance.

Additional Guidance on According to bank regulators, the accounting and
regulatory guidance

International Lending	discussed above and described in table 1 applies to
both domestic and international lending. However, bank regulators stated
that because international lending involves more risk than domestic
lending, banks that lend internationally must follow additional guidance.
The primary

Appendix III Loan Loss Allowance Guidance and Select Commercial Bank
Practices

additional risk that banks face when providing international loans is
"country risk," or the risk that economic, social, and political
conditions and events might adversely affect a bank's interests in a
country. A specific component of country risk is "transfer risk," or the
possibility that a loan may not be repaid in the currency of payment
because of restricted availability of foreign exchange in the debtor's
country.

To address country risk, banks are expected to have country risk
management methodologies in place. A 1998 study by the Interagency Country
Exposure Review Committee (the "Committee") found that all U.S. banks
conducting international lending had developed formal country risk
management programs and policies.8 The Committee also found that these
banks had formal internal country risk monitoring and reporting mechanisms
and that country risk management was typically integrated with credit risk
management. To address transfer risk, banks that lend to specific
countries must allocate additional allowances, called the Allocated
Transfer Risk Reserve.9

Transfer risk is one component of the broader concept of country risk and
the only component specifically regulated by the bank regulators. The
International Lending Supervision Act of 1983 required banks to set up an
allocated allowance for assets subject to transfer risk, and the banking
regulators accordingly published regulations implementing the requirement.
The Committee is responsible for providing an assessment of the degree of
transfer risk in cross-border and cross-currency exposure of U.S. banks
and sets the minimum amount of the allocated transfer risk reserve.10 The
Committee bases its assessments and ratings on information collected from
a number of sources, including country analysis prepared by economists at
the Federal Reserve Bank of New York and discussions with U.S. banks.11

8The Committee is composed of representatives of the OCC, FDIC, and FRB.

9The allocated transfer risk reserve is a specific allowance that is
created by a charge to current income. The allocated transfer risk reserve
is separate from the loan loss allowance and is deducted from gross loans
and leases. As far as financial statement reporting, FRB officials told us
that the allocated transfer risk reserve normally appears as part of the
loan loss allowance, that is, it is not identified separate of the
allowance.

10The Committee meets three times a year to review countries to which U.S.
banks have had an aggregate exposure of $1 billion or more for at least
two consecutive quarters.

11Two of the three banks that we spoke with stated that they were included
in the Committee's process.

 Appendix III Loan Loss Allowance Guidance and Select Commercial Bank Practices

Bank regulators emphasized that the Committee's transfer risk ratings are
primarily a supervisory tool and should not replace a bank's own country
risk analysis process. FRB officials also told us that the allocated
transfer risk reserve is "narrowly prescribed" in that it applies only to
a small number of countries. U.S. commercial bank lending is primarily
domestic, and the international lending that is conducted by banks is
concentrated in the G-10 countries12 and Switzerland. The FRB officials
stated that only approximately 30 U.S. banks-a small portion of the total
number of banks in the United States-receive allocated transfer risk
reserve statements.

Regulatory Guidance on Portfolio Segmentation and Risk Ratings

In addition to loan loss allowance guidance, banks must also follow
regulatory guidance regarding loan portfolio segmentation and risk
classification. Segmenting the bank's loan portfolio into groups of loans
with similar characteristics, such as risk classification, past-due
status, type of loan and industry, or the existence of collateral, is the
first step in calculating the loan loss allowance for the SFAS 5 portion.
Regulatory guidance states that banks may segment their loan portfolios
into as many components as practical. Bank regulators do not prescribe the
way that banks should segment their loans; however, regulatory guidance
states that loan segmentations should be separately analyzed and provided
for in the loan loss allowance. Bank regulators do provide guidance on
risk classification, a characteristic by which loan portfolios can be
segmented. Table 2 provides definitions of the risk classifications, which
are shared by all of the banking regulatory agencies.

12The G-10 countries (in addition to the United States) are Belgium,
Canada, France, Germany, Italy, Japan, Luxembourg, the Netherlands,
Sweden, and the United Kingdom.

Appendix III Loan Loss Allowance Guidance and Select Commercial Bank
Practices

                Table 2: Regulatory Agencies' Risk Rating Scale

Risk category Definition

Pass	A pass asset presents no inherent loss (no formal regulatory
definition exists for "pass" credits).

Special mention	A special mention asset has potential weaknesses that
deserve management's close attention. If left uncorrected, these potential
weaknesses may result in deterioration of repayment prospects for the
asset or in the institution's credit position at some future date. Special
mention assets are not adversely classified and do not expose an
institution to sufficient risk to warrant adverse classification.

Substandard	A substandard asset is inadequately protected by the current
sound worth and paying capacity of the obligor or of the collateral
pledged, if any. Assets so classified must have a well-defined weakness,
or weaknesses, that jeopardize the liquidation of the debt. They are
characterized by the distinct possibility that the bank will sustain some
loss if the deficiencies are not corrected.

Doubtful	An asset classified doubtful has all the weaknesses inherent in
one classified substandard with the added characteristic that the
weaknesses make collection or liquidation in full, on the basis of
currently existing facts, conditions, and values, highly questionable and
improbable.

Loss	Assets classified loss are considered uncollectible and of such
little value that their continuance as bankable assets is not warranted.
This classification does not mean that the asset has absolutely no
recovery or salvage value, but rather that it is not practical or
desirable to defer writing off this basically worthless asset even though
partial recovery may be effected in the future.

Source: OCC guidance.

Bank regulatory guidance states that a bank's rating system should reflect
the complexity of its lending activities and the overall level of risk
involved; the guidance also states that no single credit risk rating
system is ideal for every bank. Large banks typically require
sophisticated rating systems with multiple rating grades within the above
broad risk classifications. One bank regulator stated that some banks
might have a 10-point rating system based on the risk classifications,
whereas other banks may have up to 25 different ratings.

 Appendix III Loan Loss Allowance Guidance and Select Commercial Bank Practices

Regulatory Agencies For the past several years, financial regulatory
agencies and accounting

organizations have updated and continued reviewing U.S. private sector and
Accounting accounting standards and regulatory guidance governing the loan
loss Organizations Have allowance.

Been Reviewing Loan

Loss Allowance

Guidance

Regulatory Agencies Supplement Existing Guidance

In the late-1990s, Securities and Exchange Commission staff noted in their
normal reviews of filings by financial institutions, including banks, that
there were inconsistencies between the disclosures about the credit
quality of registrant's loan portfolios and the changes in the loan loss
allowances reported in the financial statements. The Securities and
Exchange Commission staff's review was aimed at determining whether the
institutions were complying with GAAP for loan loss allowances. Securities
and Exchange Commission staff was concerned, as were some of the bank
regulators, that financial institutions were (1) not using procedural
discipline in developing loan loss allowance estimates; (2) not
documenting their evaluation of loan credit quality or their measurement
of loan impairment; or (3) not providing clear disclosure, in the
financial statements and management's discussion and analysis, about the
provisioning process and allowance analysis.

As a result of the Securities and Exchange Commission staff's review of
filings, one bank restated its financial results to reflect a reduction in
its loan loss allowance. According to a bank regulator, although the
credit quality of the bank's loan portfolio was increasing, its loan loss
allowance was not decreasing.

Financial regulatory agencies issued additional guidance on the loan loss
allowance. In March 1999, the FDIC, FRB, OCC, Office of Thrift
Supervision, and Securities and Exchange Commission issued a joint letter
to financial institutions on the loan loss allowance, in which they agreed
to establish a joint working group to study the loan loss allowance and
provide improved guidance focusing on appropriate methodologies and
supporting documentation and enhanced disclosures. In 2001, the Federal
Financial Institutions Examination Council finalized and issued its Policy
Statement on Allowance for Loan and Lease Losses Methodology and
Documentation, which supplements existing regulatory guidance. The

 Appendix III Loan Loss Allowance Guidance and Select Commercial Bank Practices

Policy Statement was intended to provide further guidance on the design
and implementation of loan loss allowance methodologies and supporting
documentation practices. Securities and Exchange Commission staff issued
parallel guidance on this topic for public companies in Staff Accounting
Bulletin No. 102.

Accounting Guidance for the Loan Loss Allowance is Under Review

FASB is charged with establishing authoritative private sector accounting
principles for financial reporting.13 These accounting principles (GAAP)
are promulgated primarily through the SFAS issued by FASB. In the past,
the American Institute of Certified Public Accountants (the "Institute")
has issued industry and auditing guides to provide accounting
implementation guidance, subject to clearance by FASB.14

The Institute organized a loan loss task force with observers from the
OCC, Securities and Exchange Commission and FASB on accounting for loan
losses to "narrow the boundaries" of what is acceptable under GAAP. The
Institute's exposure draft of a proposed Statement of Position, "Allowance
for Credit Losses," was released for public comment in June 2003 and
discussed the following: the distinction between current and future
losses; how to reconcile acceptable methods for measuring loss incurred
for specific loans versus pools of loans that are collectively evaluated;
disclosure requirements; and the appropriate use of observable data in the
loan loss allowance calculation. As discussed in the March 2004 Update on
Accounting for Loan and Lease Losses, the proposed Statement of Position
raised concerns among the banking regulators and other members

13Since 1973, FASB has been the designated private sector organization
responsible for establishing standards of financial accounting and
reporting, which govern the preparation of private sector financial
statements. FASB accounting standards are recognized as authoritative by
the Securities and Exchange Commission and the Institute.

14The Securities and Exchange Commission has statutory authority to
establish accounting principles but, as a matter of policy, it generally
has relied on FASB to provide leadership in establishing and improving
accounting principles and standards. The Securities and Exchange
Commission issued a Policy Statement on April 25, 2003 recognizing FASB as
a designated accounting standard setter.

 Appendix III Loan Loss Allowance Guidance and Select Commercial Bank Practices

of the financial community who commented on the draft. In January 2004,
after review of these comment letters, the Institute decided to proceed
only with guidance to improve disclosures.15

Reviewed Banks Follow Basic Concepts in Accounting and Regulatory Guidance
but Vary in Allowance Methodologies

We spoke with three U.S. commercial banks with large lending portfolios
totaling approximately $800 billion, including large international
exposures.16 The three banks we spoke with follow the basic accounting and
regulatory concepts outlined earlier but vary in specific loan loss
allowance methodologies, including the sources of and amount of observable
data on which their allowance calculations are based. FRB, which is
conducting a study to establish "core reserving practices" among banks,
confirmed that loan loss allowance practices among banks are not
universal. The loan loss allowance varies depending on the type of lending
done by the bank and its associated levels of risk. In addition, each bank
has a unique historical loan loss experience on which their reserve
calculation is based.

Despite specific differences in loan loss allowance methodologies, banks
follow the same basic steps in determining their loan loss allowance
levels for both domestic and international loan portfolios. These steps
are illustrated in figure 7.

15The Institute's loan loss task force will evaluate existing loan loss
allowance disclosure requirements and disclosure recommendations received
through the comment process and will develop a document on list of
recommended disclosure enhancements.

16See appendix I for a further discussion of our methodology.

 Appendix III Loan Loss Allowance Guidance and Select Commercial Bank Practices

    Figure 7: Example of a Commercial Bank's Loan Loss Allowance Process for
                                Corporate Loans

Source: GAO analysis of bank information and accounting and regulatory
guidance.

Assignment of Risk Ratings	All three banks stated that their loan
portfolios are divided between their commercial and consumer businesses.
We focused on the commercial side of the loan loss allowance process of
these banks, as it was most relevant to the business of the Export-Import
Bank. Within their commercial loan portfolios (including both domestic and
international lending), regulatory guidance allows banks to segment their
loans according to various factors but risk classification is a primary
factor. The banks assign loans different risk classifications, as defined
by the bank regulators (see table 2), based on the creditworthiness of the
loan.

Appendix III Loan Loss Allowance Guidance and Select Commercial Bank
Practices

The calculation of the loan loss reserve is dependent on the risk ratings
assigned to loans. The assignment of risk ratings is based on an
assessment that includes evaluating an obligor's credit risk based on the
company or project and also on external factors, such as country risk for
international lending.

The three banks' approaches to the risk rating assignment and review
process are multilayered and performed by multiple units within the banks.
The banks we spoke with have risk management groups that are divided into
specific risk units. The groups charged with evaluating credit risk are
involved in assigning risk ratings. Ongoing analysis of the loan portfolio
is performed to ensure that risk ratings continue to be accurate. Units
within the risk management groups conduct reviews of selected loans in
their portfolios throughout the year, sometimes focusing on credits in
certain risk ratings ranges.

Factors that banks and bank examiners take into consideration when
analyzing risk in a credit exposure include industry risk; financial
indicators such as quality of cash flow, balance sheet, debt capacity, and
financial flexibility; and management. Officials at one bank told us that
they use agency ratings as benchmarks to test the reasonableness of their
internal credit risk grading system; however, the agency ratings are
considered but not specifically weighted into their rating decision.

All three banks we spoke with have committees that evaluate the country
risk levels of their lending portfolios and establish country risk ratings
and sometimes geographic exposure limits. Officials at one bank stated
that they have a formal model that assigns risk ratings to countries. The
model is based on economic, financial, social, and other factors. The
three banks incorporate information from external sources-for example,
private companies and ratings agency data-into these ratings. However, two
banks told us that, although they use external sources for data and
qualitative information, all of their analysis is internal.

Officials at one bank told us that their committee holds bimonthly
meetings and adjusts ratings monthly. Countries are placed on watch lists
when economic conditions are unstable. The watch list is based on
triggers, which include economic factors such as the pricing of debt,
exchange rates, and other political and social factors.

For international lending, the three banks factor their country risk
rating, determined internally, into the rating that they assign a loan. A
loan to a

 Appendix III Loan Loss Allowance Guidance and Select Commercial Bank Practices

foreign obligor is first rated based on the obligor's creditworthiness,
according to the banks we interviewed, then the country risk rating is
incorporated to produce an overall rating for that loan. Both the banks
and the bank regulators stated that, for international loans, the rating
assigned to a loan generally will be no better than the country risk
rating for the country in which the debtor is located. However, if a loan
is collateralized or guaranteed by a third party, the loan may receive a
rating better than the country risk rating.

FRB officials stated that Interagency Country Exposure Review Committee's
(the "Committee") country risk ratings and the allocated transfer risk
reserve requirements often lag behind the ratings of the ratings agencies
and changes already made by the banks in their reserve levels. The three
banks stated that they make their own internal judgments regarding the
allocated transfer risk reserve and can decide to have a higher allowance
than the allocated transfer risk reserve requirement, although they cannot
have a lower allowance. The banks, as did FRB officials, also stated that
the allocated transfer risk reserve is a lagging indicator and that many
specific losses have already been incurred by the time the allocated
transfer risk reserve is issued by the Committee.

Calculation of Loan Loss Allowance

Based on direction from regulatory and accounting guidance, the three
banks calculate loan loss allowances by grouping loans with similar
characteristics into pools and calculating an allowance for each pool
(which will be referred to as the "general allowance"). In other cases,
banks calculate the loan loss allowance for certain loans on an individual
basis (which will be referred to as the "specific allowance"). (Examples
of general and specific allowance calculations are illustrated in figure
7, Step B.) In calculating the loan loss allowance, banks also consider
and adjust for various factors including imprecision in the financial
models used and changing economic conditions that may affect forecasted
loan losses. 17 As with all aspects of a bank's loan loss allowance
methodology, regulatory

17The recent exposure draft Statement of Position by the American
Institute of Certified Public Accountants emphasizes that the loan loss
allowance consists of only these two components. While the term
"unallocated reserves" is commonly used in the banking industry, its
specific meaning may vary. For some, it refers to adjustments to
historical experience factors, while others believe that those adjustments
are an element of the allocated allowance for loan losses. Others believe
that unallocated refers to allowances for credit losses that are not
attributable to individual loans, referred to as the "general allowance"
in this appendix.

 Appendix III Loan Loss Allowance Guidance and Select Commercial Bank Practices

Calculation of the General Allowance

Calculation of the Specific Allowance

and accounting guidance require that support for these adjustments be well
documented.

In calculating the general allowance, loans are grouped into pools based
on similar characteristics-risk classification being a primary factor-and
collectively evaluated for impairment. The three banks we spoke with
generate expected loss factors for each loan pool by estimating such
factors as the probability of default, loss given default, and expected
exposure at default. The three banks use internal and external data to
estimate the probability of default and loss given default components. FRB
officials told us that banks tend to use external data to calculate the
probability of default and internal data to calculate the loss given
default. The practices of the three banks in our study for the most part
conformed to this view. The three banks primarily used internal data to
calculate the loss given default, sometimes validated by looking at
external data or supplemented with external data, and they primarily used
external data sources to calculate the probability of default.

With respect to the probability of default component, the banks weighted
external sources differently and used different time periods of analysis.
Officials at the three banks told us that they relied on external sources;
however, officials at one bank told us that they also internally adjusted
the data in their calculation.

The length of the historical loss experience under analysis for the banks
we interviewed varied among the loss given default components of the loan
loss allowance calculation. The three banks we spoke with used an average
of 16 years worth of data for the loss given default component.

In their specific allowance calculation, the three banks told us that they
calculate loan loss allowance for impaired loans that are larger than a
specific dollar amount on an individual basis. Among the banks with whom
we spoke, this amount ranged from hundreds of thousands of dollars to
millions of dollars. This calculation follows the guidance in SFAS 114.
For individual impaired loans, banks typically use the present value of
discounted cash flows. One bank told us that expected loss factors based
on an assessment of the loans' loss potential are determined by
consultation between loan officers and members of the risk management
group. The discounted expected future cash flows are generated using
expected loss factors, the remaining number of months that the loan is
estimated to be nonperforming, the monthly interest rate when the
obligation became nonperforming, and the gross principal balance when

 Appendix III Loan Loss Allowance Guidance and Select Commercial Bank Practices

the loan became nonperforming. The three banks periodically update their
analysis of expected loss factors. A bank regulator told us that in the
SFAS 114 calculation, banks may develop best-, base-, and worst-case
scenarios in order to make their best estimate.

The three banks pool loans that are for less than the aforementioned
dollar amount threshold and estimate losses for the loan pools. The
calculation follows guidance in SFAS 5. Two of the banks we spoke with
estimate the loss factors used in this calculation based on internal
statistical studies of historical loss experience.

Review of the Loan Loss Allowance

The three banks we spoke with review their loan loss allowance at least
quarterly, as part of the quarterly financial disclosure statements
required by the Securities and Exchange Commission. However, the banks
told us that they review large impaired loans monthly but make allowance
decisions quarterly. The risk management groups within the three banks
have the responsibility for estimating and formulating the allowance
parameters and establishing the loan loss allowance. The recommendations
and the basis of their formulation are reviewed by senior management,
whose conclusions as to the appropriateness of the loan loss allowance, as
well as the supporting analysis, are then reviewed quarterly by the bank's
board of directors.

Appendix IV

                   Interagency Country Risk Assessment System

Following enactment of the Federal Credit Reform Act in 1990, the
Interagency Country Risk Assessment System (ICRAS)-a working group of
executive branch agencies engaged in international credit activities- was
formed to provide uniformity to the process for evaluating country risk
and estimating the program costs. The Export-Import Bank (Ex-Im Bank) uses
ICRAS ratings in determining the program costs of its sovereign financing
and as a factor in its own rating process for its nonsovereign, or
private, financing. The determination of expected loss rates under the
ICRAS system has two components: (1) the assignment of risk ratings for
particular borrowers or transactions and (2) the determination of loss
rates for each risk category. Both Ex-Im Bank and the Office of Management
and Budget (OMB) play key roles in the ICRAS process-OMB chairs ICRAS, and
Ex-Im Bank provides country risk assessments and risk rating
recommendations, which are then distributed to, and agreed on, by all the
ICRAS agencies. OMB is then responsible for determining the expected loss
rates associated with each ICRAS risk rating and maturity level.

Overview of the ICRAS Framework

ICRAS was formed to satisfy the requirement of the Federal Credit Reform
Act of 1990 that common standards for country risk assessments be
established for all U.S. government agencies and programs providing
crossborder loans, guarantees, or insurance. OMB chairs ICRAS,1 Ex-Im Bank
serves as the secretariat, and several other agencies that undertake
foreign lending serve as contributing members. Economists with Ex-Im Bank
draft country papers that examine economic, political, and institutional
variables. These papers present preliminary ratings on the
creditworthiness of sovereign and nonsovereign borrowers in a country.
These papers are sent to OMB, which distributes them to other ICRAS
agencies for comment. Occasionally, agencies make major written comments
indicating disagreement with an Ex-Im Bank-recommended rating. If the
agency and Ex-Im Bank continue to disagree after discussion, OMB schedules
a meeting of all ICRAS representatives to debate the unresolved issue(s).
If there is no disagreement on its contents, or when agreement has been
reached, the recommendations of a country paper

1The ICRAS working group is chaired by OMB and includes representatives
from the crossborder financing agencies, including Ex-Im Bank, the
Departments of Agriculture and Transportation, the Overseas Private
Investment Corporation, the Agency for International Development, and the
Defense Security Assistance Administration. Other interested government
organizations, including the Departments of Treasury, State, and Commerce;
the Federal Reserve; the Council of Economic Advisors; and the National
Security Council are also represented.

             Appendix IV Interagency Country Risk Assessment System

become binding when OMB puts into effect the recommendations of a "group"
of country papers. This occurs twice each year.

Based on the results of this interagency process, OMB publishes two risk
ratings for each country-a sovereign rating and a nonsovereign, or
private, rating. Each sovereign borrower or guarantor is rated on an
11-category scale, ranging from A through F--(or their numerical
counterparts, categories 1-11). Category 1 (or A) is the most creditworthy
and category 11 (or F--) is the least creditworthy. According to Ex-Im
Bank, four categories, A through C-, are considered to be roughly
equivalent to creditworthy private bond ratings. The bottom three
categories, F through F--, are used for countries that are insolvent or
unwilling to make payments. Categories in-between represent various
degrees of repayment difficulties. These ratings must be used in
calculating the risk subsidy charged to each agency's budget when it
undertakes a foreign transaction. Each agency is free to set its own
policies with respect to fees for different risk categories and cover
policy (which specifies the risk levels at which it will undertake new
business).

Under credit reform, OMB is responsible for determining the expected loss
rates associated with each ICRAS risk rating and maturity level. OMB
provides updated expected loss rates to the ICRAS agencies for them to use
each year in preparing budget submissions, calculating reestimates, and
allocating subsidy costs during the fiscal year.

Country Risk Assessments 	In terms of extending export credits, country
risks represent risks that threaten the repayment of obligations, apart
from the financial viability of the transaction. In general terms, the
degree of risk is measured as the product of the probability of payment
delays and the probability of subsequent nonrecovery. A payment delay is
any failure to make payments of principal or interest on original contract
terms. Nonrecovery occurs in the event of default or debt forgiveness or
when there are recurring or extended arrears.

Sovereign transactions are those that carry the full faith and credit of
the central government receiving the export credit. These would typically
include transactions guaranteed by the Central Bank, Treasury, or Ministry
of Finance. On a country-by-country basis, other institutions may also be
designated as sovereign institutions, acting on behalf of the state.
According to ICRAS documents, the ability of a country to service its
foreign debt depends on the following major factors: foreign debt service

Appendix IV Interagency Country Risk Assessment System

burden, the government's ability to acquire foreign exchange to repay
foreign obligations, macroeconomic environment, and political or social
constraints. In addition to indicators reflecting those factors, ICRAS
sovereign ratings are also based on ratings of private rating agencies and
a group of Organization for Economic Cooperation and Development member
countries, as well as information on a country's payment arrears history
with the United States and other foreign creditors.

ICRAS ratings for private transactions in a country are based on
qualitative and quantitative assessments of the depth of private sector
business activity in a country and the strength of private sector
institutions. In addition to factors related to vulnerability to foreign
exchange crises, the ratings focus on a country's banking system, legal
system, foreign exchange availability, business climate, and political
stability. They can be either higher or lower than ICRAS sovereign
ratings.

Appendix V

Credit Reform Budgeting

The Federal Credit Reform Act of 1990 required that budget authority to
cover the cost to the government of new loans and loan guarantees (or
modifications to existing credits) be provided before the credits are
made. Credit reform requirements specified a net present value cost
approach using estimates for future loan repayments and defaults as
elements of the cost to be recorded in the budget. This permits policy
makers to compare the costs of credit programs with each other and with
noncredit programs in making budget decisions.

The credit reform act defines the subsidy cost of direct loans as the
present value of disbursements-over the loan's life-by the government
(loan disbursements and other payments) minus estimated payments to the
government (repayments of principal, payments of interest, other
recoveries, and other payments). It defines the subsidy cost of loan
guarantees as the present value of cash flows from estimated payments by
the government (for defaults and delinquencies, interest rate subsidies,
and other payments) minus estimated payments to the government (for loan
origination and other fees, penalties, and recoveries).

Credit programs have a positive subsidy-that is, they lose money-when the
present value of estimated payments by the government exceeds the present
value of estimated receipts. Conversely, negative subsidy programs are
those in which the present value of estimated collections is expected to
exceed the present value of estimated payments; in other words, the
programs make money (aside from administrative expenses.)

The Federal Credit Reform Act of 1990 set up a special budget accounting
system to record the budget information necessary to implement credit
reform. It provides for three types of accounts to handle credit
transactions. The program and financing accounts are used by credit
obligations made since 1991. The program account receives appropriations
for adminstrative and subsidy costs of a credit activity and is included
in budget totals. When a direct loan or a loan guarantee is disbursed, the
program account pays the associated subsidy cost for that loan to the
financing account. The financing account, which is nonbudgetary, is used
to record the cash flow associated with loans or loan guarantees over
their

                       Appendix V Credit Reform Budgeting

lives.1 It finances loan disbursements and the payments for loan guarantee
defaults with (1) the subsidy cost payment from the program account, (2)
loans from the Treasury, and (3) collections received by the government.
Figure 8 diagrams this cash flow.

Figure 8: Program and Finance Account Budgeting for Ex-Im Bank under
Credit Reform

Source: GAO analysis of credit reform guidance.

Each year, as part of the President's budget, agencies prepare estimates
of the expected subsidy costs of new lending activity for the coming year.
Agencies are also required to reestimate this cost annually. The Office of
Management and Budget (OMB) has oversight responsibility for federal

1Nonbudgetary accounts may appear in the budget document for information
purposes but are not included in the budget totals for budget authority or
budget outlay. They do not belong in the budget, because they show only
how something is financed and do not represent the use of resources.

Appendix V Credit Reform Budgeting

credit program compliance with credit reform act requirements and also has
responsibility for approving subsidy estimates and reestimates. In
addition, for international credits extended by U.S. agencies, OMB
provides agencies with specific guidance, including estimated defaults and
recoveries by risk rating category, to be used in determining expected
losses for financing activities.

All credit programs automatically receive any additional budget authority
that may be needed to fund reestimates. Thus, for discretionary programs,
original subsidy cost estimates receive different budget treatment than
subsidy cost reestimates.2 The original estimated subsidy cost must be
appropriated as part of the annual appropriation process. However, upward
reestimates of subsidy costs are financed from permanent indefinite budget
authority and do not have to be appropriated in the annual appropriations
process.3

2Discretionary programs are those controlled through the annual
appropriations process.

3Permanent budget authority is available as the result of previously
enacted legislation and does not require new legislation for the current
year. Indefinite budget authority is budget authority of an unspecified
amount.

Appendix VI

Technical Description of OMB Model for Estimating Expected Loss of U.S.
International Credit Activities

The Office of Management and Budget (OMB) determines expected losses for
international credit activities1 through (1) a complex model that includes
two estimates of default probabilities by ratings category and a rule for
combining them and (2) an assumption about how much of the value of
defaulted credits will be recovered. The default rate estimates use a
statistical concept from finance literature that OMB terms "distance to
default." The first estimated relationship-the spread-default
relationship-is between interest rate spreads on international bonds and
historical default rates of corporate debt. The second estimated
relationship-the ratings-default relationship-is between ratings on
corporate debt and the historical default rates of that debt. Historical
corporate default data are used in estimating both relationships. The
model is structured so that the overall estimates of default for different
ratings and maturities would be expected to be close to the underlying
corporate default rates used. They will differ from the underlying
historical default rates when interest rate spreads are higher or lower
than their average over the historical period of the data used in the
analysis. In addition, available information on the model suggests that
there may be certain technical biases in the model's forecasts.

Distance to Default	OMB's modeling approach uses a mathematical concept
called "distance to default," a concept used in some finance models, which
is a statistical representation of the safety of a credit. The statistical
variable has an inverse relationship with default probability-the larger
the distance to default, the smaller the probability of default. OMB's
model, in common with many models in academic finance journals, assumes
that changes in this variable follow a normal statistical distribution,
with a mean of zero, and that changes occur randomly with each time
period. Using the assumption of a normal distribution, and given an
estimated standard deviation, each distance to default implies a time
pattern of annual default rates. Distance to default is estimated by
finding the default cost implied by each distance to default and matching
that cost to the prices at which bonds of a given rating are trading.

Two forms of distance to default are used in the modeling effort. "Actual
distance to default" relates to the actual probabilities of default. "Risk

1This description is based on our analysis of information that OMB
provided about its current methodology as well as discussions with key OMB
officials. We did not review any formal documentation of the methodology.

 Appendix VI Technical Description of OMB Model for Estimating Expected Loss of
                      U.S. International Credit Activities

neutral distance to default," which is related to interest rate spreads,
refers to default rates (and recovery rates on defaulted credits) that are
consistent with observed interest rates, assuming that interest rate
spreads are attributed only to expected default costs. Finance theory
attributes the difference between actual and risk-neutral distance to
default to components of the interest rate beyond those that are related
purely to default. For example, if lenders are risk averse, rather than
risk neutral, they may need to be compensated with more than $1 of extra
interest to bear a risk of loss that may, on average, be $1, but that may
in some cases be substantially more.

Given OMB's estimated standard deviation of 3.79,2 a default rate of 25
percent for a 1-year bond implies an actual distance to default of 2.57.
This can be calculated from a standard normal distribution table. Thus,
for a given maturity, risk-free rate of interest, and standard deviation,
knowledge of any of the following factors-spread, risk-neutral distance to
default, or time pattern of default probabilities-allows the calculation
of the other two factors.

Spread-Default Relationship	The spread-default relationship is an
estimated relationship between interest rate spreads on international
bonds and historical default rates of corporate debt, by rating and
maturity. The relationship is structured so that its estimated default
rates will be close to the historical default rates used when observed
spreads are near their average levels and higher (or lower) than the
historical default rates when spreads are higher (or lower) than average.

The spread-default relationship is estimated with a regression that uses
monthly observations on about 400 sovereign bonds and historical default
rates on corporate bonds from Moody's Investors Service. The dependent
variable (the spread-related variable) is the risk-neutral distance to
default, which is calculated as a function of the monthly interest rate
spreads on the bonds in the sample. The independent variables are (1) the
actual distance to default in historical data (the default-related
variable), which is calculated for each rating and maturity as a function
of the historical corporate default rates used, and (2) the remaining
maturity of each bond.

2This is the standard deviation for the first year of the forecast. OMB's
methodology for estimating the standard deviation is discussed later in
this appendix.

Appendix VI Technical Description of OMB Model for Estimating
Expected Loss of U.S. International Credit Activities

The data used for the spread-related variable in the regression, the
riskneutral distance to default, are Bloomberg's monthly observations on
foreign sovereign bonds, denominated in U.S. dollars and issued in 1987 or
later.3 The spread on each monthly observation was calculated and
transformed into an implied distance to default to be predicted by the
regression.

The key independent variable, based on a security's rating, was calculated
as follows: ratings from Moody's and two other private ratings firms,
Standard & Poor's and Fitch Ratings, were linked to each monthly
observation. The average rating4 was calculated and used to link each
observation to an independent variable, the actual distance to default,
calculated as a function of historical default rates obtained from
Moody's. The remaining maturity of each bond, the second independent
variable, was also taken from the Bloomberg data for each monthly
observation.

This spread-related variable, risk-neutral distance to default, is
calculated by taking the spread on a bond of a given maturity and
converting it to a risk-neutral expected loss. Specifically, the
calculation determines the difference between the present value of the
payments of the bond, assuming that the bond does not default, and the
market price of the bond implied by the bond's yield. The risk-neutral
expected loss is turned into a risk-neutral expected default by solving an
equation that relates expected loss to expected default rate. This
equation calculates the present value of the losses implied by a series of
default probabilities, where defaults are converted into a series of
dollar losses by multiplying by a constant loss

3These bonds constitute the universe of foreign sovereign bonds from
Bloomberg from 1987 or later. According to OMB, observations where the
difference between the price bid and the price asked for the bond (the
bid-ask spread) of more than 10 basis points (for fiscal year 2003) or 40
basis points (for fiscal year 2004) were eliminated because these
observations may have been for illiquid instruments. This resulted in
removing about half of the observations because of large differences
between the bid and asked price for the bonds. This left 2,184 monthly
observations.

4Rating categories are those of the nationally recognized statistical
rating organizations, which are developed by the private sector and are
widely available in financial reporting. OMB translates these ratings
organizations' ratings categories into ICRAS categories.

Appendix VI Technical Description of OMB Model for Estimating
Expected Loss of U.S. International Credit Activities

rate5 and the losses are discounted by the prevailing risk-free interest
rate. Thus, a given standard deviation and a mean "distance to default"
will generate a time pattern of default rates. This mean is chosen so that
the present value of the implied dollar losses equals the risk-neutral
expected loss.

The default-related independent variable, actual distance to default, is
calculated from Moody's data on corporate defaults. Two Moody's tables
showing cumulative defaults by risk rating category and maturity were
used, one for 1920-1999, and another for 1983-1999. The tables were
combined into one table with a default rate for each combination, using
the larger of the two default rates for each rating/maturity category.
Missing table entries, or reversals (such as a higher-rated category
having a higher default rate than the next lowest category) were handled
by averaging table entries. A calculation similar to that for the
dependent variable is made, finding a mean distance to default for each
Moody's rating category that will generate a time pattern of defaults
similar to that in the Moody's tables.

Estimation of the regression produces the following parameters:

Risk-neutral distance to default = -0.26 -0.0074 * maturity + 0.73 *
actual

distance to default (Spread-related variable) (Default-related variable)

The above relationship is then inverted to produce a forecast of the
defaultrelated variable, based on the value of the spread-related
variable, resulting in the following equation:

Actual distance to default = (0.26/0.73) + (0.0074/0.73) * maturity +
(1/0.73) * risk-neutral distance to default

An autoregressive parameter is estimated from the residuals of the above
regression. This parameter is used to estimate a set of weights for
combining distance to default estimates. For every observed month, each
bond has a spread and maturity-hence, a predicted actual distance to
default. The predicted actual distance to default for each bond/month is

5OMB used an "investor loss rate" for this calculation, which OMB said
differed from the U.S. government loss rates (or recovery rates) it used
to transform its final estimates of default rates into expected loss
rates. According to OMB, the investor loss rate it used was based on the
market prices of F- -(F double minus) credits.

 Appendix VI Technical Description of OMB Model for Estimating Expected Loss of
                      U.S. International Credit Activities

averaged to produce an estimated distance to default for that bond. The
weights, derived from the autoregressive parameter, are used to construct
the weighted average. The weights are calculated so that more recent
months have more weight when taking the average.

The actual distance to default predicted by this regression depends on the
interest spread on each bond relative to the average spread for its rating
category in the Bloomberg data used in the analysis. If the spread on a
particular bond is larger than the historical average spread in the
database, then the predicted actual distance to default will be smaller
than the historical average.6 This would imply that the projected defaults
will be larger than the historical average, because projected defaults
move inversely with distance to default. Because this part of the OMB
model bases default risk on a mixture of both current spreads and past
spreads, default risk estimates will change more slowly than will the
market assessment of risk, as reflected in changes in interest rate
spreads.

Rating-Default Relationship	The relationship between ratings on corporate
debt and the historical default rates of that debt is estimated using the
Moody's corporate default tables described above. The relationship is
structured so that it predicts cumulative default rates by ratings
category and maturity that are almost exactly the same as those in the
combined Moody's tables. As with the spread-default relationship, it is
assumed that distance to default is a normally distributed variable whose
mean and standard deviation corresponds to a pattern of defaults over
time. A mean for each rating category is estimated, along with a common
standard deviation for all rating categories, that minimizes the sum of
squared errors between the cumulative default rates predicted by the means
and standard deviation and the actual data contained in the Moody's
corporate default database. A different standard deviation is estimated
for the first year than for subsequent years. This allows the actual
distance to default for any given bond in a rating category to differ from
the average distance to default for all bonds within a rating category, in
addition to allowing a bond's distance to default to change over time.

6In regression analysis, when the independent variables are at their mean
values, the dependent variable will be at its mean value. Conversely, if a
bond is near the average of the spreads observed in the data, then the
predicted actual distance to default will be close to the average observed
in the Moody's data.

 Appendix VI Technical Description of OMB Model for Estimating Expected Loss of
                      U.S. International Credit Activities

Aggregating the Estimates	The estimated actual distances to default for
each bond from the spreaddefault relationship are averaged together so
that there is one estimated distance to default for each rating category.
The estimated mean distance to default for each rating category obtained
from the spread data is then combined with the estimated mean distance to
default from the rating data. A Bayesian (type of statistical) weighting
scheme is used, giving more weight to the spread-default relationship.
According to OMB, weights vary by rating category, but generally a weight
of about two-thirds is given to the spread-default relationship and a
weight of about one-third is given to the rating-default relationship.

The result is a single actual distance to default number for each rating
category. This average value, combined with the common estimated standard
deviation for all ratings, is used to estimate annual default rates for
each rating category. An illustration of how spread changes can affect
OMB's final default estimates is shown in figure 9.

Figure 9: Illustration of How Spread Changes Can Affect the Final Expected
Default Estimates

                                  Source: GAO.

 Appendix VI Technical Description of OMB Model for Estimating Expected Loss of
                      U.S. International Credit Activities

Recovery Rates	To derive expected loss rates for each risk and maturity
category from the expected default rates generated by the model, OMB uses
an assumption about the percentage of defaulted credits that will be
recovered. According to OMB, a common recovery rate of 17 percent was used
for fiscal year 2003, a common recovery rate of 12 percent was used for
fiscal year 2004, and a common recovery rate of 9 percent was used for
fiscal year 2005.

Observations on Potential Technical Limitations of the Model

Available information on the model suggests several potential technical
limitations, including the following:

o 	The independent variable in the regression, actual distance to default,
may be measured with error. The model assumes that a particular
observation may have a distance to default that is different from the
average implied by the rating category. Additionally, the distance to
default implied by the rating category does not change over time, while
risk may change over time, even within a rating category. Measurement
error in an independent variable generally results in a downward bias in
the coefficient for that variable.7 When the estimated relationship is
reversed so that spreads are used to predict default rates, as in the OMB
model, this bias will affect the projected default rates.

o 	As noted, the actual distance to default implied by a rating category
remains constant over time, while the risk neutral distance to default,
implied by the interest rate spreads on bonds, changes over time. Thus,
the regression uses the relationship between spreads and defaults across
rating categories to produce an estimated coefficient. This coefficient is
then used to estimate default probabilities for a given rating category,
which change over time. The supporting documentation for the model does
not demonstrate a correspondence between changes in default probability
over time within a rating category and changes in default probability
across rating categories.

o 	A regression is designed to predict the dependent variable in such a
way that the squared errors in the prediction of the dependent variable
are

7See Peter Kennedy, A Guide to Econometrics, 4th ed. (Cambridge, MA: The
MIT Press, 1998), 140-142 and 148. In the case of one independent
variable, measurement error always leads to a downward bias. With more
than one independent variable, the analysis more complicated. See M. Levi,
"Errors in Variables Bias in the Presence of Correctly Measured
Variables," Econometrica vol. 41, #5 (September 1973).

Appendix VI Technical Description of OMB Model for Estimating
Expected Loss of U.S. International Credit Activities

minimized. Using the regression to predict the risk-neutral distance to
default and then inverting the estimated relationship to predict actual
distance to default may result in greater errors in the projected
distances to default than estimating the regression with actual distance
to default as the dependent variable.

o 	The relationships between risk-neutral distance to default and the two
independent variables-actual distance to default and maturity-may not be
linear. If this is the case, then spreads might provide an adequate
forecast of default probabilities near the means of the Bloomberg data set
used in the regression but not for values of spreads that depart from the
mean spread in the regression data. This issue could be important for the
reliability of estimates for credits with ratings several categories below
the average in the Bloomberg data. With sufficient data, the potential for
quantitatively important nonlinearities can be assessed by estimating
alternative specifications, such as including the squares and
cross-products of the independent variables.

Appendix VII

Comparison of OMB Default Probabilities for Fiscal Years 2004 and 2005
with Corporate Default Rates Used in OMB Model

In fiscal year 2003, the Office of Management and Budget (OMB) introduced
its current methodology for estimating the expected loss rates of
international financing provided by U.S. credit agencies. This methodology
is used to estimate loss rates for 8 of the 11 risk-rating categories
established by the Interagency Country Risk Assessment System (ICRAS).

OMB's methodology includes two components that are used to estimate
default probabilities by ICRAS rating category. One component uses default
rates for corporate bonds published in 2000 by a nationally recognized
private rating agency, Moody's Investors Service, to calculate the
probability that ICRAS agency borrowers will default. It estimates default
probabilities for each ICRAS rating category by using one or more
underlying Moody's risk category. The other component uses data on
interest rate differences, or spreads, to vertically adjust the Moody's
corporate default rates by rating category when interest rate spreads are
unusually high or low relative to average spreads in that rating category.
Once it has determined default probabilities by ICRAS rating category, the
methodology applies a recovery rate assumption to derive expected loss
rates by rating category.

We compared the default probabilities underlying OMB's fiscal year 2004
and 2005 expected loss rates for ICRAS categories 1 through 8 with the
Moody's corporate default data that OMB used in estimating these rates. We
determined that the OMB default probabilities were lower for each ICRAS
rating category in both fiscal years than were the underlying Moody's
default rates. Figures 10 through 17 compare OMB's default probabilities
for fiscal years 2004 and 2005 in a given ICRAS rating category with the
Moody's corporate default rates used in OMB's model that correspond to
each rating category. The figures show that the OMB default rates were
generally similar in fiscal years 2004 and 2005, with somewhat lower rates
in 2005 for certain ICRAS categories.

Appendix VII Comparison of OMB Default Probabilities for
Fiscal Years 2004 and 2005 with Corporate Default Rates Used in OMB Model

Figure 10: Comparison of OMB Default Probabilities for Fiscal Years 2004
and 2005 for ICRAS Category 1 with Moody's Corporate Default Rates Used in
OMB Model

Percentage default rate or probability

50 1.5

45 1.2

40

.9

35

.6 30

25 .3

20 0 12345678 15

10

5

0 12345678 Years to maturity

Corporate default rates used in OMB model

FY 2005 OMB default probability

FY 2004 OMB default probability Source: GAO analysis of OMB data.

Appendix VII Comparison of OMB Default Probabilities for
Fiscal Years 2004 and 2005 with Corporate Default Rates Used in OMB Model

Figure 11: Comparison of OMB Default Probabilities for Fiscal Years 2004
and 2005

for ICRAS Category 2 with Moody's Corporate Default Rates Used in OMB
Model

Percentage default rate or probability

50

45

40

35

30

25

20

15

10

5

0 12345678 Years to maturity

Corporate default rates used in OMB model

FY 2005 OMB default probability

FY 2004 OMB default probability Source: GAO analysis of OMB data.

Appendix VII Comparison of OMB Default Probabilities for
Fiscal Years 2004 and 2005 with Corporate Default Rates Used in OMB Model

Figure 12: Comparison of OMB Default Probabilities for Fiscal Years 2004
and 2005

for ICRAS Category 3 with Moody's Corporate Default Rates Used in OMB
Model

Percentage default rate or probability

50

45

40

35

30

25

20

15

10

5

0 12345678 Years to maturity

Corporate default rates used in OMB model

FY 2005 OMB default probability

FY 2004 OMB default probability Source: GAO analysis of OMB data.

Appendix VII Comparison of OMB Default Probabilities for
Fiscal Years 2004 and 2005 with Corporate Default Rates Used in OMB Model

Figure 13: Comparison of OMB Default Probabilities for Fiscal Years 2004
and 2005

for ICRAS Category 4 with Moody's Corporate Default Rates Used in OMB
Model

Percentage default rate or probability

50

45

40

35

30

25

20

15

10

5

0 12345678 Years to maturity

Corporate default rates used in OMB model

FY 2005 OMB default probability

FY 2004 OMB default probability Source: GAO analysis of OMB data.

Appendix VII Comparison of OMB Default Probabilities for
Fiscal Years 2004 and 2005 with Corporate Default Rates Used in OMB Model

Figure 14: Comparison of OMB Default Probabilities for Fiscal Years 2004
and 2005

for ICRAS Category 5 with Moody's Corporate Default Rates Used in OMB
Model

Percentage default rate or probability

50

45

40

35

30

25

20

15

10

5

0 12345678 Years to maturity

Corporate default rates used in OMB model

FY 2005 OMB default probability

FY 2004 OMB default probability Source: GAO analysis of OMB data.

Appendix VII Comparison of OMB Default Probabilities for
Fiscal Years 2004 and 2005 with Corporate Default Rates Used in OMB Model

Figure 15: Comparison of OMB Default Probabilities for Fiscal Years 2004
and 2005

for ICRAS Category 6 with Moody's Corporate Default Rates Used in OMB
Model

Percentage default rate or probability

50

45

40

35

30

25

20

15

10

5

0 12345678

Years to maturity

Corporate default rates used in OMB model

FY 2005 OMB default probability

FY 2004 OMB default probability Source: GAO analysis of OMB data.

Appendix VII Comparison of OMB Default Probabilities for
Fiscal Years 2004 and 2005 with Corporate Default Rates Used in OMB Model

Figure 16: Comparison of OMB Default Probabilities for Fiscal Years 2004
and 2005

for ICRAS Category 7 with Moody's Corporate Default Rates Used in OMB
Model

Percentage default rate or probability

50

45

40

35

30

25

20

15

10

5

0 12345678

Years to maturity

Corporate default rates used in OMB model

FY 2005 OMB default probability

FY 2004 OMB default probability Source: GAO analysis of OMB data.

Appendix VII Comparison of OMB Default Probabilities for
Fiscal Years 2004 and 2005 with Corporate Default Rates Used in OMB Model

Figure 17: Comparison of OMB Default Probabilities for Fiscal Years 2004
and 2005

for ICRAS Category 8 with Moody's Corporate Default Rates Used in OMB
Model

Percentage default rate or probability

50

45

40

35

30

25

20

15

10

5

0 12345678

Years to maturity

Corporate default rates used in OMB model

FY 2005 OMB default probability

FY 2004 OMB default probability Source: GAO analysis of OMB data.

Appendix VIII

Trends in Interagency Country Risk Assessment System Expected Loss Rates

Through the Interagency Country Risk Assessment System (ICRAS), the Office
of Management and Budget (OMB) annually provides expected loss rates to
ICRAS agencies to use in preparing their budget submissions and subsidy
cost estimates. The expected loss rates, issued for each of the 11 ICRAS
risk categories, have changed in percentage terms over time. Figure 18
shows the trends in expected loss rates for ICRAS categories 1 through 8
for credits of 8-year maturity, expressed in present value terms, for
fiscal years 1997 through 2005.1

1During the period analyzed, the format in which OMB presented expected
loss rates varied. For fiscal years 1997 through 2002, OMB presented risk
premiums for ICRAS categories 1 through 8, which were grouped into several
maturity bands. From these premiums, an expected loss rate could be
derived. Beginning in fiscal year 2003, OMB changed its presentation into
expected loss rates for ICRAS categories 1 through 8, across different
maturities. To show trends over time, we converted the risk premiums into
expected loss rates.

                                 Appendix VIII
    Trends in Interagency Country Risk Assessment System Expected Loss Rates

Percentage expected loss

45

40

35

30

25

20

15

10

5

0 1997 1998 1999 2000 2001 2002 2003 2004 2005 Fiscal year

ICRAS score

Source: GAO analysis of OMB expected loss rates.

Note: The present values were calculated for credits of 8-year maturity
using OMB's credit subsidy calculator based on a discount rate of 5
percent in each fiscal year.

Appendix IX

Comments from the Office of Management and Budget

Appendix IX
Comments from the Office of Management
and Budget

Appendix X

                     GAO Contacts and Staff Acknowledgments

GAO Contacts	Celia Thomas, 202-512-8987 Shirley Brothwell, 202-512-3865

Staff 	In addition to those individuals named above, Allison Abrams,
Nathan Anderson, Dan Blair, Patrick Dynes, Reid Lowe, Ernie Jackson,
Austin

Acknowledgments 	Kelly, Bruce Kutnick, Berel Spivack, and Roger Stoltz
made major contributions to this report.

GAO's Mission	The Government Accountability Office, the audit, evaluation
and investigative arm of Congress, exists to support Congress in meeting
its constitutional responsibilities and to help improve the performance
and accountability of the federal government for the American people. GAO
examines the use of public funds; evaluates federal programs and policies;
and provides analyses, recommendations, and other assistance to help
Congress make informed oversight, policy, and funding decisions. GAO's
commitment to good government is reflected in its core values of
accountability, integrity, and reliability.

Obtaining Copies of The fastest and easiest way to obtain copies of GAO
documents at no cost

is through GAO's Web site (www.gao.gov). Each weekday, GAO postsGAO
Reports and newly released reports, testimony, and correspondence on its
Web site. To Testimony have GAO e-mail you a list of newly posted products
every afternoon, go to

www.gao.gov and select "Subscribe to Updates."

Order by Mail or Phone	The first copy of each printed report is free.
Additional copies are $2 each. A check or money order should be made out
to the Superintendent of Documents. GAO also accepts VISA and Mastercard.
Orders for 100 or more copies mailed to a single address are discounted 25
percent. Orders should be sent to:

U.S. Government Accountability Office 441 G Street NW, Room LM Washington,
D.C. 20548

To order by Phone:	Voice: (202) 512-6000 TDD: (202) 512-2537 Fax: (202)
512-6061

To Report Fraud, Contact:
Waste, and Abuse in Web site: www.gao.gov/fraudnet/fraudnet.htm

E-mail: [email protected] Federal Programs Automated answering system:
(800) 424-5454 or (202) 512-7470

Congressional	Gloria Jarmon, Managing Director, [email protected] (202)
512-4400 U.S. Government Accountability Office, 441 G Street NW, Room 7125

Relations Washington, D.C. 20548

Public Affairs	Jeff Nelligan, Managing Director, [email protected] (202)
512-4800 U.S. Government Accountability Office, 441 G Street NW, Room 7149
Washington, D.C. 20548

                               Presorted Standard
                              Postage & Fees Paid
                                      GAO
                                Permit No. GI00

United States Government Accountability Office Washington, D.C. 20548-0001

Official Business Penalty for Private Use $300

Address Service Requested
*** End of document. ***