A Guide to IMF Stress Testing
Chapter

Chapter 33. A Forward-Looking Macroprudential Stress Test for U.S. Banks

Author(s):
Li Ong
Published Date:
December 2014
Share
  • ShareShare
Show Summary Details
Author(s)
Geoffrey N. Keim and Andrea M. Maechler This chapter is an abridged version of IMF Country Report 10/244 (IMF, 2010). The authors wish to thank Sergei Antoshin, Ivailo Arsov, Peter Dattels, Dale Gray, Charles Kramer, David Robinson, Erica Tsounta, and participants at seminars held at the Board of Governors of the Federal Reserve, the U.S. Department of the Treasury, and the IMF for their helpful comments and guidance.

This chapter presents a forward-looking macroprudential stress test methodology for 53 U.S. bank holding companies to assess how changes in the operating environment may affect the medium-term health of individual banks and the system as a whole. By forecasting key elements of the banks’ balance sheets, it is possible to capture the interactions between banks’ earnings potential, their losses and capital positions, and their ability to support economic growth through lending. Scenario analysis then is used to explore banks’ resilience to different stresses, from worse-than-anticipated macroeconomic conditions to more targeted shocks, such as large-scale up-front principal write-down on their impaired residential mortgages. Our results indicate that banks’ balance sheets may not be as strong as their capital buffers would suggest.

Method Summary

Method Summary
OverviewThe forward-looking macroprudential stress test is an exercise that models the impact of a projected path for key macroeconomic and financial variables on banks’ capital buffers and other balance sheet positions, including their ability to provide credit.
ApplicationThe method has proven particularly useful in assessing banks’ ability to withstand a severe but plausible adverse scenario over a forecast period, while maintaining an adequate capital base to provide credit to the real economy.
Nature of approachAccounting based (balance sheet and financial statements), forward looking (over a two- to four-year horizon), and with macro-financial linkages.
Data requirementsBank-specific balance sheet data and financial statements, historical based projections of key economic and financial variables.
StrengthsIf needed, the model can be run solely on publicly available data or supplemented by supervisory information where available; the flexibility of the framework allows capturing a wide range of shocks and scenarios.
WeaknessesThe framework is based on low-frequency balance sheet data; it does not capture default dependencies and other cross-firm asset correlations and requires judgment about tail-risk calibrations; it tends to focus on the first-round impact, although the GDP forecast accounts for banks’ asset growth in an iterative way.
ToolStandard econometrics package.

The Excel spreadsheet macro is available in the toolkit, which is on the companion CD and at www.elibrary.imf.org/stress-test-toolkit.

Contact author: G. Keim.

To assess the resilience of the banking system to changes in the U.S. macroeconomic environment, the IMF Financial Sector Assessment Program (FSAP) team conducted a forward-looking balance sheet—based analysis. This analysis is similar to the Supervisory Capital Assessment Program (SCAP) exercise, conducted by the authorities in early 2009, in the sense that it forecasts banks’ capital needs into the future based on particular macroeconomic projections.1 In contrast to the SCAP, however, the present exercise is based entirely on publicly available information as of end-March 2010.2 This approach is also related to the capital adequacy analysis presented in the Global Financial Stability Report (GFSR) with regard to aggregate loss estimates; however, it differs from the GFSR in its bank-specific “bottom-up” focus on earnings, losses, and capital positions.

The exercise covers 53 bank holding companies (BHCs), representing 85.2 percent of all bank BHC assets, including a number of regional and smaller banks with less widely tracked information (Table 33.1).3 To capture differences across sizes and business strategies, we grouped the sample into seven subcategories, namely, the “top 4” (accounting for 46.7 percent of sample assets); the 2 former investment banks (10.3 percent of sample assets); 9 regional banks (7.7 percent of sample assets); 3 processing banks (2.7 percent of sample assets); 3 consumer banks (3.2 percent of sample assets); 21 “small” banks (3.4 percent of sample assets); and 11 foreign banks (11.1 percent of sample assets).4 The rest of the system was grouped into a residual category, which accounted for 14.8 percent of sample assets.5

Table 33.1List of Institutions, Total Assets as of End-March 2010
InstitutionNumber of InstitutionsTotal Assets (in billions)In Percent of Sample
Total system5416,483.9100.0
Top 4 banks47,702.346.7
Bank of America Corporation12,340.714.2
JPMorgan Chase & Co.12,135.813.0
Citigroup, Inc.12,002.212.1
Wells Fargo & Company11,223.67.4
Investment banks21,700.410.3
Goldman Sachs Group, Inc.1880.75.3
Morgan Stanley1819.75.0
Bear Stearns
Lehman
Regional banks91,273.97.7
PNC Financial Services Group, Inc.1265.41.6
U.S. Bancorp1282.41.7
SunTrust Banks, Inc.1171.81.0
BB&T Corporation1163.71.0
Regions Financial Corporation1137.30.8
Fifth Third Bancorp1112.70.7
KeyCorp195.30.6
Popular, Inc.133.80.2
W Holding Company, Inc.1111.50.1
Washington Mutual
Processing banks450.22.7
Bank of New York Mellon Corporation1221.01.3
State Street Corporation1152.90.9
Northern Trust Corporation176.30.5
Consumer banks3522.43.2
Capital One Financial Corporation1200.71.2
American Express Company1142.30.9
Ally Financial21179.41.1
CIT
Ameriprise
Small banks21564.93.4
Comerica Incorporated157.20.3
Marshall & Ilsley Corporation156.60.3
Zions Bancorporation151.70.3
Huntington Bancshares Incorporated151.90.3
Synovus Financial Corp.132.40.2
New York Community Bancorp, Inc.142.40.3
First Horizon National Corporation125.90.2
BancorpSouth, Inc.113.20.1
Associated Banc-Corp123.10.1
BOK Financial Corporation123.50.1
First BanCorp.118.90.1
Webster Financial Corporation118.00.1
Commerce Bancshares, Inc.118.00.1
TCF Financial Corporation118.20.1
First Citizens BancShares, Inc.121.20.1
First National of Nebraska, Inc.116.60.1
City National Corporation120.10.1
Fulton Financial Corporation116.40.1
New York Private Bank & Trust Corporation113.10.1
Susquehanna Bancshares, Inc.113.80.1
South Financial Group, Inc.112.40.1
Foreign banks111,822.011.1
M&T Bank Corporation168.40.4
Harris Financial Corp.165.50.4
BancWest Corporation175.20.5
UnionBanCal Corporation185.50.5
Barclays Group US, Inc.1427.82.6
BBVA USA Bancshares, Inc.165.20.4
HSBC North America Holdings, Inc.1345.42.1
RBC Bancorporation (USA)126.20.2
Taunus Corporation1364.12.2
TD Banknorth, Inc.1154.70.9
Citizens Financial Group, Inc.1144.00.9
Other12,447.814.8
Source: SNL Financial.

The purpose of the stress tests is to assess the soundness of BHCs, including under “worse-than-anticipated” macroeconomic conditions. By forecasting key elements of the banks’ balance sheets, the exercise captures the interactions between banks’ earnings potential, their capital positions, and their ability to absorb losses. All else equal, it also provides some insights into banks’ ability to support the economic recovery through lending. Last, the analysis can be used to explore the vulnerability of BHCs to a wide range of stresses, from broad macro scenarios to more targeted shocks.

The results of the stress test should be interpreted with caution. These are subject to uncertainty from a number of sources, including the specification of our statistical models, the level of bank-level detail, the possibility of even more severe and unforeseen events, and the validity of our assumptions on banks’ future business practices. Many of these factors would be present under any forecasting exercise. Moreover, historical correlations that were observed in the past may not be indicative of the relationship that can be expected going forward in light of the substantial economic shock experienced during the crisis, the many and varied associated policy responses that have followed, and the more recent vulnerabilities in Europe. Thus, when interpreting the results, it is important to appreciate that the results are point estimates, and there is uncertainty around them, which is not quantified. Nonetheless, where appropriate, standard tests and alternative scenarios provide some indication of the sensitivity of the results to underlying assumptions.

The results contain both upside and downside risks. The largest downside risk pertains to banks’ earnings outlook, which is assumed to recover around their 1990–99 historical average, with an average annualized return on assets (ROA) of 2 percent for the 2010–14 sample periods. These estimates are likely to be on the high side, as they do not incorporate the impact of the upcoming financial regulatory reforms, both domestically and internationally, banks’ greater risk retention in the absence of a return to precrisis securitization levels, and lower credit growth in line with the relatively weak economic outlook. Another important downside risk surrounds loss estimates, particularly due to the large uncertainties regarding the new phenomenon of strategic defaults in the case of “underwater” mortgages and banks’ recovery rates, given the potentially long-lasting depressed collateral values. On the upside, banks’ ability to raise private capital or reduce their dividend policy could significantly strengthen their capacity to absorb losses and support economic growth when demand recovers.6

The assessment of BHCs’ capital adequacy over the forecast period employed several capital measures. To assess the quality of capital, while allowing comparability both across countries and to the SCAP, three capital metrics were used; (1) the ratio of Tier 1 capital to risk-weighted assets with 6 and 8 percent thresholds; (2) the SCAP’s Tier 1 common capital/risk-weighted assets ratio with 4 and 6 percent thresholds; and (3) a tangible common equity to tangible assets ratio with 4 and 6 percent thresholds. The thresholds were not ambitious relative to historical norms (SCAP institutions maintained an average 10 percent Tier 1 capital ratio during the crisis, and their Tier 1 common capital ratio was on average 7.4 percent over 1997-2007) (Figure 33.1).7

Figure 33.1Capital Position of Bank Holding Companies, 1997-2010 (in percent of risk-weighted assets)

Source: SNL Financial.

Note: bn = billion; SCAP = Supervisory Capital Assessment Program.

The exercise spans over a seven-year horizon. It used realized quarterly data from end-2007 to end-March 2010 and produced quarterly forward-projections until end-2014. Most of the bank-specific data came from the publicly available Y-9C reports that bank holding companies file with the Fed and were obtained from SNL Financial’s database. The regulatory data were augmented further with Securities and Exchange Commission data for nonbanks before their BHC conversion in 2008, Bloomberg for capital-raising measures and securities write-downs, and the U.S. Treasury Department’s Web site http://www.FinancialStability.gov for Troubled Asset Relief Program (TARP) repayments and dividends.

1. Baseline Scenario

A. Framework

The analysis projected firms’ net revenues, losses, and balance sheet expansion to assess banks’ potential capital shortfalls over a five-year period. A bank’s capital shortfall (if any) was computed based on its lowest capital position over the horizon. Consideration was given to firm-specific differences in earnings and losses, based on portfolio composition and historical performance. An attempt also was made to capture a number of specific postcrisis factors that would affect BHCs’ asset growth, including banks’ efforts to deleverage and de-risk their balance sheets, the new Financial Accounting Standard 166/167 accounting rules requiring banks to on-board certain assets previously held off-balance-sheet, and greater risk retention due to impaired securitization. Moreover, the calculations also incorporated firms’ ability to accumulate tax assets in loss-making quarters that could be used to offset future tax liabilities.

Macro-financial linkages are built into the stress test framework. By modeling how macroeconomic variables have influenced historically the behavior of specific financial variables, it is possible to link a particular macroeconomic path to financial sector developments and their related impact on financial firms’ capital position. The nominal GDP growth forecast, for example, drives asset growth; loan loss rates reflect movements in the path for real GDP, real consumption, the unemployment rate, and the output gap. Other macroeconomic variables critical for the loss estimates, such as lending standards and house prices, are forecasted separately (Appendix I). Although the link between macroeconomic variables and financial variables is the centerpiece of this exercise, it should be noted that there is no universal, consensus view as to how these variables should relate to each other and that judgmental adjustments may be needed.

The baseline scenario was taken from the IMF’s April 2010 World Economic Outlook (WEO). In particular, the output gap closed over the medium term from a negative level in 2009, with the unemployment rate remaining elevated (above 8 percent) until end-2011 before dropping to 5.5 percent by end-2014. Real GDP growth was expected to peak at 3.1 percent in 2010 and to stabilize around 2.5 percent by 2012. House prices were expected to rise over the forecast horizon, albeit at a very slow pace (peaking at 4.1 percent in 2011) (Appendix I).

B. Underlying assumptions and forecasting methodology

Five categories of loan charge-off rates were estimated on an industry-wide basis from regression analysis (Appendix II). These include losses on commercial real estate (CRE) loans, residential real estate (RRE) loans, commercial and industrial (C&I) loans, and consumer loans (Figure 33.2).8 To capture historical cross-firm differences, BHC-specific charge-off rates were projected for each type of loan, given by CORi,qj, where i denotes firms (with I being the industry average), j indexes the loan type, and q denotes time. The forecasted rates were computed recursively taking as their base the previous quarter’s value, to which the change in the industrywide charge-off rate for that class of loan ΔCOR¯I,qj was applied:

Figure 33.2Baseline Scenario: Quarterly Loss Profiles, 2007-14 (in billions of dollars)

Sources: Authors; and SNL Financial.

Note: C&I = commercial and industrial; CRE = commercial real estate; RRE = residential real estate; SCAP = Supervisory Capital Assessment Program.

No adjustment was made to account for much stricter underwriting standards post-2009. In practice, stricter lending standards should help reduce future loss estimates and particularly their sensitivity to adverse shocks. In the near term, this omission should not play a large role in the context of falling credit growth rates. For the outer years of the forecast, however, it could lead to an upward bias in the loss estimates.

Only limited account was taken of the mergers and acquisitions that took place in 2008 among several large banks. In the SCAP exercise, adjustments for losses already taken on impaired loans acquired through mergers (i.e., purchase accounting adjustments) reduced estimated losses by $64 billion. No such adjustment was made in the current exercise, largely because of the difficulty of assessing the performance of the acquired loans relative to expectations in the absence of detailed loan-by-loan information. Furthermore, it is reasonable to assume that the impact of such purchasing accounting adjustments would diminish over time, including through amortization.

Minor adjustments were made for the investment banks that converted to bank holding companies in late 2008. Because of their recent conversion, focusing on historical prudential data would have put an unreasonably large weight on their (poor) performance during the crisis. In the case of Morgan Stanley, the calculations omit the company’s abnormally high loan loss rates during the fourth quarter of 2008 from the moving average used to forecast its future losses. The company’s earnings path was raised by adjusting it to the estimated average of the fixed effects of the top six firms (Figure 33.3).9

Figure 33.3Estimated Bank-Specific Effects and Group Averages in Return on Asset Regressions

Source: Authors.

The results suggest that credit risk is likely to remain a source of concern for some time, with some loss rates not peaking before mid-2011 (Table 33.2). Some of the highest loss rates may have already peaked, such as consumer loans, which reached 6.5 percent at end-January 2010, and RRE, which rose to 2.7 percent at end-2009. Losses on CRE loans, however, are expected to continue to rise until mid-2011, as they are generally expected to peak later than in the RRE sector. The high loss rate on consumer credit reflects partly the aggressive charge-off policy of the consumer banks in an effort to clean their balance sheets, possibly in anticipation of a pickup in demand following the sharp credit line contractions since 2008. The relatively low C&I loan loss rate, which peaked at 2.6 percent at end-September 2009, reflects the relatively healthy financial position of the corporate sector, which was able to either use its cash buffers to pay down debt or benefit from advantageous refinancing terms in the corporate bond market.

Table 33.2Peaks for Loan Loss Charge-off Rates, 2009–14 (in percent)
Baseline ScenarioAdverse ScenarioAlternative Scenario
SectorMax.PeriodMax.PeriodMax.Period
RRE2.72009: Q43.42011: Q43.52012: Q1
Consumer6.52010: Q16.52010: Q16.52010: Q1
CRE3.42011: Q24.62011: Q35.12011: Q4
C&I2.62009: Q32.62009: Q32.62009: Q3
Other3.42009: Q43.82011: Q23.62011: Q3
Sources: Authors; Bloomberg; and SNL Financial.Note: C&I = commercial and industrial; Cons = construction; CRE = commercial real estate; RRE = residential real estate.

While showing signs of stabilization, losses on CRE loans are projected to remain high over the forecast horizon. Unless commercial property prices start recovering from their 30 percent fall since mid-2006, banks are likely to face heavy losses, given the large volume of underwater mortgage borrowers and upcoming adjustable-rate mortgage resets. Furthermore, the weak economy continues to hammer rents and occupancy rates in many markets, with negative consequences for defaults, foreclosures, and losses. The Congressional Oversight Panel (2010), for example, estimated that about $1.4 trillion in loans will mature in 2010-14, nearly half of which are already seriously delinquent (90 days or more past due) or “underwater” (with a loan value exceeding the property value).

Real estate loan quality is vulnerable to downside risks. Although the loss rate on RRE loans is expected to decline going forward, the rising gap between delinquencies and foreclosures suggests a large volume of pent-up supply of houses for sale through foreclosures, which could put further downward pressure on house prices and exacerbate strategic defaults for some time to come. As of end-March 2010, seriously delinquent loans (90 days or more overdue) accounted for 5 percent of total mortgages, in contrast to actual foreclosures, which accounted for only 1.25 percent of total loans. Loan modifications could help mitigate the risk of a pickup in foreclosures; however, redefault may be high, in which case modifications simply could postpone losses further into the future.

In the baseline, total cumulative loan losses are expected to reach $802 billion by end-2014 ($592 billion for SCAP firms). This represents a 6.5 percent cumulative loss for 2010–11 (12.3 percent for 2010–14) (Table 33.3). Although the two-year loss rates are below the 9.1 percent 2009–10 loss rate assumed in the SCAP stress test, they amount to an annual average of 3.3 percent for 2010–11 and 2.5 percent for 2010–14. Consumer banks face the largest two-year loss rate, followed by the top four banks. Small and regional banks face lower but still material loss rates (9.7 and 9.4 percent, respectively), reflecting their heavy exposure concentration to CRE.

Table 33.3Cumulative Loss Rates, 2010-14 (in percent)
InstitutionBaselineAdverse ScenarioAlternative Scenario
Loan LossesTotal LossesLoan LossesTotal LossesLoan LossesTotal Losses
2010-112010-142010-112010-142010-112010-142010-112010-142010-112010-142010-112010-14
Total6.412.16.412.17.715.98.717.47.615.28.215.8
Top four8.115.38.115.39.819.810.821.39.719.510.320.2
Regional bank5.19.45.19.46.313.47.014.56.312.46.712.8
Consumer banks8.918.78.918.710.722.211.523.410.220.310.620.8
Small bank5.59.75.59.76.513.37.014.16.512.06.812.3
Foreign banks6.310.96.310.97.715.38.716.77.614.88.215.4
Annual average3.22.43.22.43.83.24.43.53.83.04.13.2
Source: Authors.

Source: Authors.

C. Securities write-downs

Write-downs on securities were measured as declines in market valuations based on the methodology developed and updated in a recent GFSR.10 Under the baseline, no additional securities write-down on available-for-sale securities was expected, and the framework did not envisage any shocks to marked-to-market trading account securities (Table 33.4). Since the beginning of the crisis in end-2007, BHCs reported a cumulative $385 billion of realized marked-to-market securities write-downs (not shown), relative to the $296 billion of write-downs estimated by the model. To be on the conservative side, we made no allowance for write-ups to banks’ securities holdings.

Table 33.4Securities Write-down Projections
SecuritiesEstimated HoldingsJanuary 2010

Cumulative Losses
January 2010 Cumulative

Loss Rate (Percent)
Share of Total

(Percent)
Residential mortgage1,47216611.356.2
Agency (prime conforming)49200.00.0
ABS (home and multifamily)98016617.056.2
of which: nonagency prime MBS53000.00.0
of which: ABS (CDOs, other MBS)45016637.056.2
Consumer14200.00.0
Commercial mortgage1964824.516.3
Corporate1,115171.55.6
Governments58000.00.0
Foreign975666.722.2
Total for securities6,9322966.6100.0
Sources: Authors; and Bloomberg.Note: ABS = asset-backed security; CDO = collaterized debt obligation; MBS = mortgage-backed security.

D. Earnings profiles

One of the most challenging elements of this exercise was to forecast banks’ earnings. The analysis focused on preprovision, pretax, and predividend net revenues as a percentage of total assets, henceforth referred to as ROA. The ultimate regression, which covers 53 BHCs, was run using a fixed-effects panel specification. It included a vector Xt of three macroeconomic variables (real GDP quarterly growth rate, output gap, and lagged quarterly unemployment growth rate), one bank-specific variable yit-1 (the lagged loan-to-asset ratio) to capture banks’ different business strategies, and one financial market variable, zit (the spread between the three-month London interbank offered rate [LIBOR] and the Treasury bill of similar maturity), as a proxy for financial market conditions; uit is the unit-specific residual, and £tt is the usual residual:

The model was estimated using quarterly frequency data from 1990:Q1 to 2010:Q1. The macroeconomic variables were adjusted seasonally, whereas the bank-specific and financial variables were not. For the forecast period, the estimated coefficients were applied to the forecasted explanatory variables, allowing the resulting retained earnings to feed back into total assets each quarter. No attempt was made to model subgroups of institutions, although the fixed effect from the panel regression allowed introducing bank-specific differences (Figure 33.3). Key results are shown in Table 33.5. The final model specification is highlighted in Column (4). The results for the macroeconomic regressions are presented in Columns (1) and (2), whereas the fixed-effects regressions are presented in Columns (3) to (5), and those for the single and multilevel mixed effects regressions are presented in Columns (6) and (7).

Table 33.5Summary Results for Return on Asset Regressions
Macro DataFixed-effects PanelMixed-effects Panel
(1)(2)(3)(4)(5)Bank and

Group

Level (6)
Bank Level

(7)
Loan-to-asset (lagged)0.584***1.792***0.276*0.307**0.292*0.324***0.317***
−1.06E-04−8.83E-08−0.0714−0.0433−0.05080.00E + 000.00E + 00
Log of total assets (lagged)0.0684***−1.65E-02
−6.07E-06−0.138
Real GDP quarterly growth2.4322.704**1.363**1.687**1.356**1.690***1.708***
−0.15−0.0363−0.0471−0.014−0.0449−0.00238−0.00213
3-month LIBOR to 3-month T-Bill−0.201***−0.188***−0.151***−0.150***−0.103***−0.150***−0.150***
−7.42E-08−4.54E-07−1.28E-09−1.15E-09−1.28E-050.00E + 000.00E + 00
Output gap0.0129*−0.01040.0240***0.0197***0.0204***0.0196***0.0194***
−0.051−0.225−0.000185−0.00332−0.00232−1.24E-10−1.65E-10
Unemployment quarterly growth (lagged)0.2960.03190.1380.02050.02620.02150.0225
−0.101−0.848−0.153−0.857−0.817−0.805−0.795
Dummy for 2007-08−0.0929***
−0.000193
Constant0.269***−1.925***0.698***0.393***0.389***0.447***0.396***
−2.08E-04−9.85E-05−0.00199−1.64E-04−1.63E-04−1.24E-100.00E + 00
Number of observations79793,4013,4013,4013,4013,401
Number of banks535353
R20.6500.736
Adjusted R20.6260.7140.1360.1320.139
Within = R20.1380.1340.140
Standard deviation residual error (e_it)0.1760.1760.176
Number of groups753
Random effect at bank level−1.613***−1.478***
0.00E + 000.00E + 00
Random effect at group level−1.948***
1.90E-05
Standard deviation of overall error term−1.734***−1.734***
0.00E + 000.00E + 00
Source: Authors.Notes: Independent variable is return on assets. Robust p-values indicated in italics below coefficient: *p < 0.1; **p < 0.05; ***p < 0.01. LIBOR = London interbank offered rate; T-Bill = Treasury bill.

A wide range of model specifications were tested to estimate earnings, including macroeconomic (industry-wide) models and different bank-specific panel models. Given the objective of linking the earnings forecast to various macroeconomic scenarios, the choice of explanatory variables was restricted to those that could be linked directly to the macroeconomic model used for our scenario analysis (e.g., GDP, output gap, unemployment, real consumption) or for which there was an in-house forecasting model (e.g., house prices, yields on the LIBOR, or Treasury bills) that could be linked to our scenario analysis. Real personal consumption, expenditures, growth, and house prices (both on an unadjusted and detrended basis) were found to be statistically insignificant. The lagged log of total assets (L.lnta), which was used to capture differences associated with size, was not found to be statistically significant. The inclusion of different financial market variables yielded limited differences if using different spreads, including a 10-year high yield bond spread to 10-year treasuries and a 10-year to 3-month term structure. The three-month LIBOR to Treasury bill spread was chosen for convenience, as these variables were part of our macroeconomic modeling forecast.

Given the macroprudential focus of the exercise, model specifications do not claim to be definitive in forecasting bank earnings. Instead, the analysis is designed to capture the sensitivity of banks’ earnings to changes in the macroeconomic variables used in our stress scenarios. Our estimates would not be appropriate for forecasting earnings of individual institutions or groups of institutions, given that no attempt was made to model different business lines. It is possible that the crisis and the subsequent changes in the financial landscape (e.g., new patterns of competition, impact of regulatory reform or limited securitization) have changed the underlying relationships between banks’ earnings and their determinants. Not surprisingly, there are uncertainties over the earnings outlook over the forecast period, with significantly reduced accuracy of the estimates in the outer years (Figure 33.4).11 Thus, parameter estimates may not be applicable, and the forecasts would contain a bias. Nonetheless, the estimates provide some indication of the earnings capacity of the system as a whole (with group-specific variances) and the sensitivity of these earnings to various shocks.

A range of common statistical tests were used to select the final model specification. In particular:

Figure 33.4Return on Assets, Historical and Model Forecast, 1990-2014

Sources: Authors; and SNL Financial.

  • The Akaike and Bayesian information criteria were helpful in narrowing down the choice of explanatory variables.12

  • The Hausman specification and likelihood-ratio tests were used to differentiate across model specifications.13

  • Both the Durbin-Watson d−statistic and the Breusch-Godfrey serial correlation Lagrange multiplier test indicate that serial correlation would disappear with the inclusion of bank-specific variables.14

  • The random-effects model was rejected consistently based on the Breusch-Pagan Lagrange multiplier test as an appropriate model specification.15

  • Multilevel mixed-effects panel models also were run. Dummies were introduced to account for the dramatic downturn in 2007 and 2008. They were not included in the final specification, as they did not help improve the model as the effects of the crisis would be captured through real GDP.

According to our baseline, industry-wide bank earnings would remain modest, closer to mid-1990s levels, with large variations across subgroups (Figure 33.5). Return on asset is expected to average 1.96 percent on an annualized basis over the forecast horizon. This is substantially higher than the SCAP exercise, which assumed that banks’ ROA would remain almost 15 percent below the past 20-year average for 2009–10, or around 1.6 percent on an annualized basis.

Figure 33.5Pretax, Preprovision Net Revenue of Commercial Banks, 1984-2009 (in percent of total assets)

Source: Federal Deposit Insurance Corporation, Historical Statistics on Banking. Note: ROA = return on assets.

Note: ROA = return on assets

The larger banks are expected to yield larger-than-average earnings, in line with historical experience. By category, the projected ROA for the top four bank holding companies is expected to average an annualized 2.17 percent over the forecast horizon, or around its 1990–99 historical average of 2.2 percent. By comparison, the average ROA forecast is 2.33 percent for regional banks, 1.75 percent for small banks, and 1.36 percent for foreign banks (Figure 33.6). As with the larger banks, earnings are well below the levels observed in the years immediately preceding the crisis, which were 2.37 percent for regional banks, 2.26 percent for small banks, and 1.70 percent for foreign banks. The projected ROA is slightly lower over the next two years, or 1.9 percent for the system relative to 2 percent over 2010–14. In the adverse scenario, the system is expected to average an annualized 1.6 percent until end-2012, or 1.7 percent over the forecast horizon.

Figure 33.6Baseline and Adverse Scenarios: Annualized Return on Asset, by Subgroups, 1990-2014

Sources: Authors; and SNL Financial.

Note: Pretax, preprovision, predividend net revenues to total assets.

E. Retained earnings

To estimate retained earnings, we paid close attention to firms’ tax profiles, including their ability to defer tax assets in loss-making periods.16 Broadly, the framework applied a simple 30 percent flat tax rate on banks’ corporate income. In addition, it accounted for banks’ ability to accumulate deferred tax assets (DTAs), which could be used later to pay for future tax liabilities.17 In particular, when a BHC would make losses following periods when it had taxable income, it would be allowed to carry back operating losses for two years to recover income taxes previously paid and accumulate tax benefits against future (positive) income going forward. These carrybacks are DTAs as they could be used to pay down future tax liabilities. A fraction of these DTAs would qualify as Tier 1 capital (we assumed up to 10 percent of Tier 1 capital).18 When the institution would make profits again, it would draw down its accumulated DTAs to pay for its tax liabilities, thereby boosting retained earnings and hence organic capital growth.

According to our baseline results, the large institutions would benefit materially from DTAs over the forecast horizon. Cumulative DTAs peaked at end-March 2009 at $143 billion for the system, 68 percent of which was accounted for by the top 4 institutions (Figure 33.7). By end-2014, DTAs would help reduce future tax liabilities by 82 percent. In the baseline, 21 institutions would not have to pay income tax over the sample horizon, including one of the top 4 institutions and 5 regional banks (in the adverse scenario, the number of firms would rise to 31 institutions, including 3 of the top 4 institutions).

Figure 33.7Baseline Scenario: Impact of Deferred Tax Assets, 2005–14

Sources: SNL Financial; and IMF staff calculations.

Note: Cum. = cumulative.

A straightforward dividend rule was applied to all financial firms in the baseline. In particular, when net aftertax income was positive, the calculations assumed a 5 percent annualized dividend rate for TARP preferred shares, 8 percent for other preferred shares (relative to an average of 5 percent over 1990-99), and 15 percent for common equity (relative to an average of 22 percent over 1990-99). This resulted in an 11.6 percent annualized average dividend rate for common equity and 2.6 percent annualized average dividend rate for preferred shares (Figure 33.8). This is significantly lower than historical dividend rates. In the downside risk scenarios, however, banks were not expected to pay out dividends on their common shares, in line with the assumption underlying the authorities’ SCAP exercise.

Figure 33.8Baseline and Adverse Scenarios: Historical and Projected Dividends, 1990-2014 (in percent of pre-tax income)

Sources: Authors; and SNL Financial.

Under the baseline, banks’ retained earnings would be sufficient to cover losses over the forecast period. For the industry as a whole, retained earnings (defined as preprovision pretax net revenue minus loan charge-offs, securities write-downs, taxes, and dividends) would remain positive, although low (slightly above $20 billion on average) until end-2011, at which point they would start rising (Figure 33.9). Regional banks follow a similar pattern (with average quarterly retained earnings of less than $1 billion until end-2011), while small banks face negative retained earnings until the first quarter of 2012. Over the full 2010-14 forecast horizon, retained earnings for the system would average $43 billion on a quarterly basis ($34 billion for SCAP firms, $4 billion for regional banks, and less than $1 billion for small banks).

Figure 33.9Baseline Scenario: Retained Earnings, 2007-14 (in billions of dollars)

Sources: Authors; and SNL Financial.

Note: PPNR = preprovision pretax net revenue; SCAP = Supervisory Capital Assessment Program.

F. Balance sheet expansion

Over the 2010-14 horizon, asset growth is slightly weaker than nominal GDP growth due to deleveraging (Figure 33.10). The calculations control for a number of factors that affect balance sheet expansion. In particular, weak securitization markets, which could lead banks to retain a larger-than-ordinary share of assets on their balance sheets, are predicted to add $195 billion to total system assets. The introduction of the FAS 166/167 accounting rules in 2010, which require banks to bring on balance sheet a significant amount of assets previously held off-balance-sheet, are also assumed to expand banks’ balance sheets by $375 billion.

Figure 33.10Balance Sheet Expansion (in billions of dollars)

Sources: SNL Financial; and IMF staff calculations.

Furthermore, retained earnings were added back into total assets, adding $670 billion to banks’ balance sheets over the sample horizon (64 percent of which was generated by the top 6 firms). Factors tempering growth of total system assets included asset sales, which subtracted $375 billion, and asset maturities without rollovers, which reduced assets by $496 billion. Except for retained earnings, which were estimated on a bank-by-bank basis, the balance sheet expansion factors were distributed across firms according to their share of total system assets. Total assets can be decomposed as follows:

where TAit is total assets for bank i at time t (I refers to the BHC sample); rgdp_qg is the quarterly growth rate of real GDP; BSEit and REit are, respectively, the projected balance sheet expansion and the estimated retained earnings; and cou and wdu are, respectively, the estimated loan charge-offs and securities write-downs.

The path for risk-weighted assets also was modeled carefully. Since end-2007, the ratio of risk-weighted assets to total assets has fallen by over 5 percentage points to 61 percent, the lowest point recorded since the introduction of risk-weighted assets (Figure 33.11). This falling trend reflects banks’ substantial efforts to “de-risk” their balance sheets since the onset of the crisis. However, it would not be likely for banks to maintain such a low ratio, especially as they expand their lending activities. Thus, it was assumed that, under the baseline, banks’ risk-weighted to total asset ratios would return progressively back to their 2000-2005 average by mid-2011 (the adverse scenarios assumed that the ratio would remain constant at the low end-March 2010 level).

Figure 33.11Baseline Scenario: Bank Holding Company Asset Composition, 1990-2014 (in percent)

Sources: Authors; and SNL Financial.

Note: RWA = risk-weighted assets; TA = total assets.

Furthermore, the composition of BHCs’ balance sheets was allowed to adjust to ensure maximum room for credit expansion. Broadly, the framework allowed banks’ loan portfolios to grow in proportion to their asset growth, assuming a constant loan-to-asset ratio. However, this could have materially underestimated credit growth, given banks’ record low loan-t o-asset ratio at end-March 2010. Instead, BHCs also were allowed to expand their loan portfolios by drawing down up to 5 percent of their “other assets” for eight consecutive quarters (or until “other assets” reached 20 percent of total assets). As a result, the loan-to-asset ratio was raised by 7 percentage points to 50 percent by the end of the forecast horizon, although still well below historical averages. The path for total loans can be decomposed as follows:

where TLit stands for total loans, OAit for other (nonloan, nonsecurity) assets, and a for the fraction by which securities can be substituted for loans.

G. Capital shortfall estimates

Under the baseline scenario, bank capital would be adequate on an industry-wide basis. Notwithstanding weak growth, high unemployment, and record high charge-off rates, the top four BHCs and the former broker dealers are expected to maintain a 6 percent Tier 1 common equity ratio over 201014 (bottom panel in Table 33.5). However, three SCAP institutions would require an addition of $7.4 billion in capital to maintain the same ratio. Because of high CRE exposure, four regional banks (including two SCAP institutions) may require $1.3 billion in additional capital, while seven smaller institutions would likely require an additional $6.3 billion. Subsidiaries of foreign banks, which tend to be lightly capitalized and rely on parental support, may require up to $26.3 billion. Overall, the system would require $40.5 billion in additional capital. The top four institutions would need to raise $40.4 billion in additional capital if required to maintain a 5.9 percent tangible common equity to tangible assets ratio (or 17 times leverage).

Our results suggest that weak financial institutions should be encouraged to raise capital as current conditions do not allow them to grow out of their problems. The picture of capital shortfall does not change materially when focusing on a shorter two-year forecast horizon (upper panel in Table 33.6). Overall, the capital shortfalls would affect the same institutions. This suggests that weak institutions are not able to rely on organic growth to improve their financial condition. The system as a whole could require as much as $33.6 billion in additional capital to maintain a 6 percent Tier 1 common capital ratio, most of which borne by the foreign banks ($26.3 billion).

Table 33.6Baseline Scenario: BHC Capital Needs, 2010-14 (in billions of dollars, unless otherwise noted)
ScenarioTop FourInvestmentRegionalProcessingConsumerSmallForeignOtherTotalU.S. Only
2010:Q2-2011:Q4 (cumulative)
Pretax, preprovision net revenue351.097.668.121.338.822.445.992.2737.3691.4
Loan losses276.60.251.20.926.427.740.362.3485.5445.2
Securities losses1.510.060.190.000.000.130.000.01.91.9
Taxes2.1616.845.005.970.610.41−0.800.831.031.8
Dividends18.913.74.22.13.81.22.39.856.053.7
Addition to retained earnings53.366.47.612.16.3−6.94.523.5166.8162.3
Capital injection end-2011 to reach Tier 1 capital/risk-weighted assets ratio
6 percent0.00.00.00.00.01.626.60.028.21.6
8 percent0.00.00.00.00.02.637.80.040.42.6
Number of banks requiring
injection
6 percent0000034N/A73
8 percent0000034N/A73
Tier 1 common capital/risk-weighted assets ratio1
Capital injection end-2011 to reach
4 percent0.00.00.00.00.02.715.80.018.52.7
6 percent0.00.00.50.02.24.526.30.033.67.2
Number of banks requiring
injection
4 percent0010044N/A95
6 percent0040174N/A1612
Tangible common equity/tangible assets ratio
Capital injection end-2011 to reach
4 percent (25 times leverage)0.02.70.40.02.93.432.20.041.69.5
5.9 percent (17 times leverage)40.418.34.43.27.06.953.40.0133.480.0
Number of banks requiring
injection
4 percent (25 times leverage)0120154N/A139
5.9 percent (17 times leverage)21512116N/A2822
2010:Q2-2014:Q4 (cumulative)
Pretax, preprovision net revenue895.0262.5179.756.799.555.2121.6244.01914.31792.6
Loan losses496.30.387.21.551.446.266.0111.9860.9794.9
Securities losses1.50.10.20.00.00.10.00.01.91.9
Taxes43.866.326.315.512.43.56.930.6205.4198.5
Dividends68.534.314.65.98.04.18.823.8168.2159.4
Addition to retained earnings286.2161.151.533.625.91.340.381.8681.8641.5
Capital injection at lowest point for Tier 1 capital/risk-weighted assets ratio
6 percent0.00.00.00.00.02.826.6029.52.8
8 percent0.00.00.00.00.03.837.8041.63.8
Number of banks requiring
injection
6 percent0000034N/A73
8 percent0000034N/A73
Tier 1 common capital/risk-weighted assets ratio1
4 percent0.00.00.00.03.84.115.80.023.77.9
6 percent0.00.01.30.06.66.326.30.040.514.2
Number of banks requiring
injection
4 percent0010144N/A106
6 percent0040174N/A1612
Tangible common equity/tangible assets ratio
4 percent (25 times leverage)0.02.70.60.07.74.832.20.047.915.7
5.9 percent (17 times leverage)40.418.35.33.212.19.053.50.0141.688.2
Number of banks requiring
injection
4 percent (25 times leverage)0120154N/A139
5.9 percent (17 times leverage)21512116N/A2822
Memo: Percent of total system assets46.510.48.02.82.93.511.314.6100.088.7
Sources: Authors; and SNL Financial.Note: BHC = bank holding company.

The estimated capital shortfall of foreign banks is difficult to interpret. The current exercise stresses foreign institutions in the same way as it does domestic ones as a way of assessing the broader shock absorption capacity of the U.S. banking system. In normal times, foreign holding companies tend to operate with lower capital buffers than their domestic peers, as they are not required to comply with the U.S. regulatory capital requirements, provided their parents are deemed well capitalized and well managed. Under a global adverse shock, however, it could be particularly difficult for regulators to require higher capital buffers when parent banks could be equally strained. Although the resulting retrenchment or closure of foreign banks would likely not have systemic consequences from a financial stability perspective, it may have broader macroprudential implications depending on the operations of the affected institutions.19 Since end-2007, foreign BHCs reduced their loan market share by 5 percentage points to 9 percent.

Credit growth could remain limited for some time (Figure 33.12). Although the financial system appears stable from a financial stability perspective, its relatively low level of retained earnings, combined with banks’ recent efforts to deleverage and de-risk their balance sheets, may result in limited credit expansion, even after accounting for internal asset substitution away from cash and other assets into loans. Our results suggest that, in the absence of additional capital injections, credit growth could average around 8 percent for 2010–14, which is substantially lower than historical levels. For example, credit growth rates averaged around 16.1 percent in 1993–96 (following the savings and loan crisis) and 16.8 percent in 2004–07 (after the 2002–03 recession). In the adverse scenario, the average credit growth could fall by another 2 percentage points for the forecast horizon. In reality, banks will have various ways to meet credit demand, including by raising new capital, curbing dividend rates, or managing to generate higher retained earnings than anticipated.

Figure 33.12Baseline and Adverse Scenarios: Credit Growth, 1990-2014 (year-on-year in percent)

Sources: Authors; and SNL Financial.

2. Alternative Scenarios

To test banks’ shock absorption capacity under worse-than-anticipated macroeconomic conditions, we considered downside risk scenarios. These included an adverse macroeconomic scenario and an alternative funding risk scenario. The assumed values were consistent with historical distress episodes, and the magnitudes of the shocks are broadly in the ranges analyzed in other FSAPs (details on the alternative scenario are presented in Appendix I). These scenarios also help demonstrate the sensitivity of the baseline results to underlying assumptions.

A. Adverse scenario

Under the adverse scenario, loan losses continue increasing appreciably. Residential and CRE loan losses continue to rise until 2011 (peaking at 3.4 percent and 4.6 percent, respectively), while losses on consumer and C&I loans rise further, without reaching their earlier peaks (Table 33.7). Cumulative loan losses are expected to reach $1.1 trillion for the system as a whole by end-2014, representing a 7.7 percent cumulative loss rate for 2010-11 (15.9 percent for 2010-14). In addition, the institutions with securities portfolios are also expected to write down $100 billion of marked-to-market securities, resulting in a total cumulative loss rate of 17.4 percent for 2010-14 (8.7 percent for 2010-11).

Table 33.7Adverse Scenario: BHC Capital Needs, 2010-14 (in billions of dollars, unless otherwise noted)
ScenarioTop FourInvestmentRegionalProcessingConsumerSmallForeignOtherTotalU.S. Only
2010:Q2-2011:Q4 (cumulative)
Pretax, preprovision net revenue307.588.260.718.635.519.336.080.5646.3610.3
Loan losses327.20.260.91.131.132.248.569.5570.6522.2
Securities losses32.700.176.474.052.032.475.3216.569.764.4
Taxes−2.1813.981.206.890.460.00−1.730.419.120.8
Dividends5.93.61.80.41.50.70.33.818.017.7
Addition to retained earnings−54.869.8−9.66.1−1.2−16.0−16.0−8.5−30.2−14.2
Capital injection end-2011 to reach Tier 1 capital/risk-weighted assets ratio
6 percent0.00.00.30.00.02.819.40.022.53.1
8 percent0.00.00.90.00.04.628.90.034.35.5
Number of banks requiring injection
6 percent0010044N/A95
8 percent0020055N/A127
Tier 1 common capital/risk-weighted assets ratio1
Capital injection end-2011 to reach
4 percent0.00.01.50.02.25.214.60.023.58.9
6 percent0.00.05.50.05.29.524.40.044.620.2
Number of banks requiring injection
4 percent0030164N/A1410
6 percent00501115N/A2217
Tangible common equity/tangible assets ratio
Capital injection end-2011 to reach
4 percent (25 times leverage)8.62.33.20.74.56.936.50.062.826.2
5.9 percent (17 times leverage)106.317.910.94.68.812.662.70.0223.8161.1
Number of banks requiring injection
4 percent (25 times leverage)11411115N/A2419
5.9 percent (17 times leverage)41712127N/A3427
2010:Q2-2014:Q4 (cumulative)
Pretax, preprovision net revenue770.8238.9156.149.589.646.394.9208.71654.81559.8
Loan losses633.30.4121.42.260.561.590.8143.21113.41022.6
Securities losses47.20.29.45.93.03.67.824.1101.193.3
Taxes6.359.110.113.47.01.40.83.0101.2100.4
Dividends12.67.63.60.41.50.90.47.534.434.0
Addition to retained earnings72.8171.111.827.415.9−21.0−4.432.1305.7310.1
Capital injection at lowest point for Tier 1 capital/risk-weighted assets ratio
6 percent0.00.02.80.00.09.125.5037.411.9
8 percent0.00.06.40.00.612.937.7057.519.9
Number of banks requiring injection
6 percent0030066N/A159
8 percent00401107N/A2215
Tier 1 common capital/risk-weighted assets ratio1
4 percent0.00.08.10.08.814.921.80.053.631.8
6 percent0.00.012.80.012.119.731.70.076.444.6
Number of banks requiring injection
4 percent00401104N/A1915
6 percent00501116N/A2317
Tangible common equity/tangible assets ratio
4 percent (25 times leverage)15.22.39.80.711.216.443.40.099.155.7
5.9 percent (17 times leverage)119.117.918.44.615.722.572.00.0270.3198.2
Number of banks requiring injection
4 percent (25 times leverage)11411115N/A2419
5.9 percent (17 times leverage)41712138N/A3628
Memo: Percent of total system assets46.510.48.02.82.93.511.314.6100.088.7
Sources: Authors; and SNL Financial. Note: BHC = bank holding company.

On aggregate, BHCs no longer would be able to absorb their losses through earnings in the near term. Retained earnings would remain negative until 2012 for the system as a whole and until 2014 for the smaller banks. The SCAP firms would fare slightly better with retained earnings turning positive by end-2011. Retained earnings for the system would record an average quarterly loss of $2.4 billion for 2010-11 ($1.2 billion for the regional banks and $1.8 billion for the small institutions).

Almost half of the U.S. BHCs would experience some capital shortfall under the adverse scenario (Table 33.7). U.S. BHCs would require a total of $31.8 billion capital to maintain a 4 percent Tier 1 common capital ratio until end-2014 ($53.6 billion including the foreign BHCs). In particular, 4 regional banks would require $8.1 billion, 10 smaller institutions another $14.9 billion. Three SCAP banks would face a shortfall of $14.5 billion. One of the top 4 institutions would need to raise $15.2 billion to maintain a 4 percent tangible common equity to tangible assets ratio by end-2014. Over the 2010-11 horizon, 10 U.S. BHCs (including 2 SCAP institutions) would be expected to face a capital shortfall of $8.9 billion to maintain a 4 percent Tier 1 common capital ratio.

B. Alternative scenario

Rollover risk warrants careful surveillance. Market liquidity risks appear to have declined, thanks to effective and powerful policy response by the authorities during the crisis. Financial institutions, however, remain vulnerable to the potential risk posed by the large volume of CRE loans that are expected to mature between 2010 and 2014 (many of which with negative equity) when real estate prices may not have yet recovered, together with the rising stock of seriously delinquent mortgages on banks’ balance sheets.

The alternative scenario tests the sensitivity of banks’ capital shortfall estimates to a further small deterioration in the CRE sectors in 2010–11. Under this scenario, the macroeconomic conditions are broadly similar to those in the adverse scenario for the first two years but return faster to the baseline beyond 2011 (Appendix I). CRE prices are expected to fall by another 8 percent by end-2012 (as opposed to 3.3 percent in the adverse), while house prices are expected to fall by 4.1 percent in 2010 and another 2.6 percent in 2011. Banks’ assumed difficulty in rolling over maturing debt leads to higher losses on CRE loans, which peak at 5.1 percent at end-2011.

Our results suggest that, except for banks already heavily exposed to CRE, macroeconomic conditions are currently the key determinant to banks’ financial soundness. Broadly, our results under the alternative scenario (Table 33.8) are not materially different from those in the adverse scenario. Overall, 14 U.S. BHCs would require $20.5 billion capital to maintain a 4 percent Tier 1 common capital ratio over the 2010–14 period ($7.4 billion over the 2010-11 period). This suggests that banks that are heavily exposed to the CRE losses will find it difficult to earn their way out of their problems under worse-than-expected macroeconomic conditions, but this result is not highly sensitive to a further small deterioration in real estate prices or recovery rates on delinquent real estate loans. Clearly, a broader shock to banks’ funding conditions that leads, for example, to a substantive rise in short-term spreads would likely have a more dramatic impact on banks’ earnings and hence on their ability to absorb losses.

Table 33.8Alternative Scenario: BHC Capital Needs, 2014-14 (in billions of dollars, unless otherwise noted)
ScenarioTop FourInvestmentRegionalProcessingConsumerSmallForeignOtherTotalU.S. Only
2010:Q2-2011:Q4 (cumulative)
Pretax, preprovision net revenue313.389.361.719.035.919.737.682.6659.1621.5
Loan losses326.50.261.01.129.832.348.069.5568.3520.3
Securities losses18.910.123.702.261.131.442.979.239.736.8
Taxes−2.1814.351.617.350.460.08−1.730.420.422.1
Dividends6.43.61.90.41.50.70.35.520.320.0
Addition to retained earnings−35.070.6−6.37.81.4−14.7−11.5−0.811.422.9
Capital injection end-2011 to reach
Tier 1 capital/risk-weighted assets ratio
6 percent0.00.00.20.00.02.518.40.021.12.7
8 percent0.00.00.70.00.04.026.30.031.14.7
Number of banks requiring injection
6 percent0010034N/A84
8 percent0020055N/A127
Tier 1 common capital/risk-weighted assets ratio1
Capital injection end-2011 to reach
4 percent0.00.00.90.01.74.813.40.020.87.4
6 percent0.00.04.30.04.78.422.00.039.417.4
Number of banks requiring injection
4 percent0030153N/A129
6 percent00401115N/A2116
Tangible common equity/tangible assets ratio
Capital injection end-2011 to reach
4 percent (25 times leverage)0.02.52.30.43.85.934.10.049.014.9
5.9 percent (17 times leverage)78.418.08.74.38.111.559.70.0188.6128.9
Number of banks requiring injection
4 percent (25 times leverage)01311115N/A2217
5.9 percent (17 times leverage)41612127N/A3326
2010:Q2-2014:Q4 (cumulative)
Pretax, preprovision net revenue812.6247.3163.552.392.749.4106.4223.61747.71641.4
Loan losses625.30.4112.92.155.756.088.2128.41068.8980.7
Securities losses21.10.14.12.51.31.63.310.444.641.2
Taxes8.961.714.615.49.22.13.216.1131.3128.0
Dividends14.57.64.10.41.51.50.49.239.138.7
Addition to retained earnings144.2177.127.931.723.3−11.611.660.8464.9453.3
Capital injection at lowest point for
Tier 1 capital/risk-weighted assets ratio
6 percent0.00.00.80.00.05.921.4028.16.7
8 percent0.00.02.70.00.08.031.9042.710.8
Number of banks requiring injection
6 percent0010055N/A116
8 percent0030075N/A1510
Tier 1 common capital/risk-weighted assets ratio1
4 percent0.00.04.00.06.410.017.80.038.320.5
6 percent0.00.08.20.09.714.527.60.060.032.4
Number of banks requiring injection
4 percent0040194N/A1814
6 percent00401115N/A2116
Tangible common equity/tangible assets ratio
4 percent (25 times leverage)0.02.55.90.49.011.639.30.068.629.3
5.9 percent (17 times leverage)79.318.012.64.313.517.365.70.0210.5144.8
Number of banks requiring injection
4 percent (25 times leverage)01411115N/A2318
5.9 percent (17 times leverage)41612128N/A3426
Memo: Percent of total system assets46.510.48.02.82.93.511.314.6100.088.7
Sources: Authors; and SNL Financial.Note: BHC = bank holding company.

3. Conclusion

The results confirm that BHCs’ asset quality and capital positions are closely interlinked with developments in the housing sector and the broader macroeconomy. There is much uncertainty about the shape and height of the loss profiles, although they are expected to be a drag on retained earnings and credit growth. Identified fragilities in regional and smaller institutions do not appear systemic but could hamper economic recovery in local communities with broader repercussions on bank loss rates. To mitigate this risk, the authorities intend to allocate $30 billion of TARP money to community banks. Another potential vulnerability is the low capitalization of foreign-owned BHCs.

Despite the significant improvement in BHCs’ capital buffers, our results suggest that additional capital may be needed to create room for meaningful credit growth. Since the crisis, BHCs have managed to almost double their holdings of “high-quality” capital. Nonetheless, the current combination of record low risk-weighted to total asset ratio, outlook for a protracted period of high loss profiles, limited risk transfer through securitization, and general recognition that financial institutions need to hold higher capital buffers than in precrisis means that banks’ balance sheets may not be as strong as their capital buffers would suggest.

Appendix I. Stress Test Scenarios and Shocks for the U.S. Financial Sector Assessment Program

The stress tests in the U.S. Financial Sector Assessment Programs (FSAPs) included a baseline scenario and alternative scenarios. These included an adverse macroeconomic scenario and an alternative funding risk scenario.20 The assumed values were consistent with historical distress episodes, and the magnitudes of the shocks are broadly in the ranges analyzed in other FSAPs.

Baseline Scenario

The baseline was the scenario from the IMF’s April 2010 World Economic Outlook update. The output gap closes over the medium term from a negative level in 2009, while inflation is well anchored and stabilizes at about 2¼ percent. Ten-year government bond yields continue to rise moderately from 3.3 percent to 6.6 percent by 2015, reflecting the increasing government debt-to-GDP ratio.

Adverse Scenario

The adverse scenario was generated using a simple closed-economy business cycle model for the United States, with standard monetary channels (Taylor rule and nominal rigidities) and fiscal channels (a fiscal rule and a link between the real interest rate and government debt).21 The scenario was calibrated to illustrate the combined impact of four adverse shocks: (1) a sizable and persistent shock to the growth rate of potential output, reflecting continued difficulties in the financial system and very weak investment; (2) an additional short-term demand shock, reflecting high unemployment, weak credit, and continued fall in housing prices; (3) further near-term fiscal stimulus to support near-term growth; and (4) rising inflation expectations, reflecting concerns over medium-term fiscal risks and renewed higher oil prices.

Reflecting this combination of shocks, the output gap falls by another 2.3 percentage points in 2011 relative to the baseline, and the unemployment rate remains close to 10 percent in 2010-11. House prices fall by another 2.2 percent in 2010 and 2.1 percent in 2011, before starting a modest upward trend in 2012. Reflecting the weaker macroeconomic environment, banks’ annualized return on assets falls slightly over the forecast horizon. Inflation and government bond yields rise, and the government debt-to-GDP ratio increased by almost 10 percentage points compared with the baseline in 2013 (Table 33.9).

Table 33.9Macroeconomic Assumptions (percent change, unless otherwise noted)
Scenario20102011201220132014
Baseline
Real GDP3.12.62.42.52.4
Real personal consumption expenditures2.42.12.02.02.0
Nominal GDP3.94.04.24.44.3
Output gap (percent)−2.0−1.0−0.6−0.3−0.1
Unemployment rate (percent)9.88.97.05.85.5
Case-Shiller 10-city house prices2.12.02.92.51.5
Spread of 3-month LIBOR to 3-month T-Bill0.20.40.40.40.4
Return on assets (annualized; percent)1.71.81.81.81.9
Adverse
Real GDP2.3−0.80.8−1.72.60.22.60.12.2−0.2
Real personal consumption expenditures1.9−0.60.6−1.61.6−0.41.5−0.51.3−0.6
Nominal GDP3.8−0.23.4−0.74.1−0.24.60.24.50.2
Output gap (percent)−3.0−1.0−3.3−2.3−2.1−1.5−1.1−0.8−0.6−0.4
Unemployment rate (percent)10.00.29.91.08.91.97.71.96.91.5
Case-Shiller 10-city house prices−2.2−4.3−2.1−4.12.2−0.72.50.01.80.2
Spread of 3-month LIBOR to 3-month T-Bill0.30.10.60.30.70.30.60.20.60.2
Return on assets (annualized; percent)1.6−0.21.4−0.51.5−0.41.6−0.31.7−0.3
Alternative funding risk
Real GDP2.4−0.60.8−1.81.6−0.82.50.02.40.0
Real personal consumption expenditures2.3−0.20.2−1.90.0−2.01.3−0.71.7−0.2
Nominal GDP3.3−0.71.9−2.13.1−1.14.40.04.40.1
Output gap (percent)−3.3−1.3−2.6−1.6−0.8−0.3−0.30.0−0.10.0
Unemployment rate (percent)10.60.89.91.07.20.15.80.05.50.0
Case-Shiller 10-city house prices−4.1−6.1−2.6−6.73.10.32.40.01.50.0
Spread of 3-month LIBOR to 3-month T-Bill0.40.10.50.10.50.10.40.00.40.0
Return on assets (annualized; percent)1.5−0.31.6−0.31.7−0.21.8−0.11.9−0.1
Source: Authors.Note: Shaded numbers denote deviations from baseline. LIBOR = London interbank offered rate.

Alternative Scenario

The alternative funding risk scenario was conducted to test banks’ resilience to a further small deterioration in the real estate sectors, including difficulties in rolling over their commercial real estate (CRE) maturing debt and continuing to accumulate seriously delinquent mortgages on their balance sheets. The fact that nearly half of the $1.4 trillion in CRE loans expected to mature between 2010 and 2014 have negative equity (see Congressional Oversight Panel, 2010), together with the rising stock of seriously delinquent mortgages (many of which are “underwater”), suggests that banks could face difficulties in refinancing a large volume of loans and face larger real estate loan losses if economic conditions do not improve and real estate prices do not rebound.

Under this scenario, the output gap falls more sharply in 2010 (3.3 percent relative to 3 percent), and the unemployment rate rises faster (10.6 percent relative to 10 percent). House prices are expected to fall by 4.1 percent in 2010 and another 2.6 percent in 2011, while CRE prices fall by another 8 percent by end-2012 (as opposed to 3.3 percent in the adverse scenario). Short-term market spreads react slightly more than under the adverse scenario in 2010 but return faster to the baseline in the outer years, allowing banks to earn higher profits over the forecast horizon. Importantly, banks’ assumed difficulty in rolling over maturing debt leads to higher losses on CRE loans, which peak at 5.1 percent at end-2011.

Single-Factor Shocks

In addition to the scenarios, a range of single-factor shocks were employed to examine resilience of the financial system with respect to individual risk factors. The calibration of these shocks was based on long-term U.S. historical data as well as experience from other countries.

Appendix II. Industry-Wide Loan Loss Projections for Bank Holding Companies

Charge-off rates for different loan types are modeled as dependent on a set of economic and financial variables.22 The general methodology was developed initially and subsequently refined in the context of the systemwide capital shortfall estimates presented in the Global Financial Stability Report (see IMF, 2008b, 2009a, for a description of the methodology). This appendix outlines the methodology for forecasting bank charge-off rates, lending standards, and house prices.

In order to better capture future turning points in the charge-off patterns, levels and log levels (rather than growth rates) were used for the explanatory variables. Because a decline in bank lending standards indicates a slower rate of tightening, the use of cumulative net balances for lending standards was warranted (referred to as “cumulative lending standards” hereafter). This was to reflect that, for example, charge-offs can continue to rise despite a slowdown in house price declines and a deceleration in the pace of tightening in lending standards. Similarly, to capture deterioration in economic conditions amid a slowdown in negative growth rates, we used the output gap (which can be viewed as the detrended level of real GDP), instead of GDP growth.

The underlying historical data on loan loss rates and lending standards were obtained from the Fed, while macroeconomic and financial data came from Haver Analytics. Where available, forecast data were taken from the World Economic Outlook (WEO). Housing prices and lending standards were modeled separately. The sample was composed of quarterly data from 1991 to 2009 so as to incorporate the last two recessions.

Charge-Off Rates

To deal with nonstationarity in the variables, the empirical Bayesian approach was employed. The estimation was carried out by running 10,000 Markov Chain Monte Carlo simulations using the Gibbs sampler package WinBUGS (Lunn and others, 2000). Convergence was obtained within 1,000 burn-in runs. The estimated coefficients in the resented equations were statistically significant at 5 percent. Lending standards were particular to each type of loan.

Real estate charge-off rates

Charge-off rates for real estate loans followed a two-step approach. First, the percentage of loans that would become delinquent was estimated. Second, the “hazard rate” or the transition rate was estimated to capture the percentage of delinquent loans that would transition into actual charge-offs. The exact specification for commercial and residential charge-off rates looked as follows:

Commercial real estate charge-off rate (C_CRE)

The first term D_CREt is the delinquency rate, which is modeled as a function of the cumulative lending standards LS_CRE and CRE prices CP:

The second term is the transition rate, which is modeled as a function of the cumulative lending standards:

Residential real estate charge-off rate (C_CRE)

The delinquency rate is a function of the cumulative lending standards and level of RRE prices:

The transition rate is a function of the cumulative lending standards and unemployment rate:

Consumer loans charge-off rate (C_CONS)

The consumer loans charge-off rate is modeled as a function of the cumulative lending standards and the output gap:

Commercial and industrial loans charge-off rate (C_CI)

The commercial and industrial loans charge-off rate is modeled as a function of the detrended cumulative lending standards and the output gap:

Lending Standards

The lending standards reported in the Board of Governors of the Federal Reserve System’s (“Fed”) Senior Loan Officer Opinion Survey of Bank Lending Practices (SLOOS) appear as an independent variable in each of the charge-off and delinquency models. The respondents to the survey, which is usually conducted quarterly, include large U.S. commercial banks and large branches and agencies of foreign banks. For each loan category, the respondents report whether over the past three months they have tightened, loosened, or maintained the criteria used to determine whether a borrower is creditworthy. The Fed reports these responses as the percentage of respondents indicating that their bank had tightened standards less the percentage indicating that their bank had loosened standards for a number of different kinds of loans. Whenever these measures are positive, it is generally more difficult for borrowers to obtain loans.

To ground forecasts of these lending standards, we estimated four vector autoregression models. The variables in each system included real GDP growth, CPI inflation, the change in corporate bond spreads, the change in the Federal Funds Target Rate, oil prices, and the growth of total bank loans. Further, each model contained one loan category’s lending standards. Given that SLOOS asks banks to report on their lending standards over the previous three months, we lagged the lending standards one quarter so that the change in standards would be contemporaneous with the changes in the other variables. We obtained forecasts for each class of lending standards from the estimated systems of equations, treating the macro variables as exogenous, similar to the approach discussed by Swiston (2008). Output from these models estimated with detrended lending standards augurs a loosening in lending standards for all types of loans over the forecast horizon (Figure 33.13). These results are in line both with recent developments in the SLOOS and with the baseline scenario, which calls for continued macroeconomic growth and financial stabilization. Nevertheless, to align the forecasts with the historical behavior of lending standards in previous business cycles, we adjusted model-based projections with judgmental forecasts.

Figure 33.13Senior Loan Officer Opinion Survey: Lending Standards (de-trended net percentage indicating tightening standards)

Source: Federal Reserve.

Note: C&I = commercial and industrial; CONS = consumer loans; CRE = commercial real estate; RRE = residential real estate.

House Prices

Prices for residential real estate (RRE) and commercial real estate (CRE) enter the equations used to forecast RRE and CRE loan delinquency rates. Neither the WEO nor the macro simulation models used in the scenario analysis contain a forecast for prices of housing or CRE. Thus, to generate the forecasts required for the delinquency rate models, we used a model for house prices and an observation about the house prices–CRE prices to generate the CRE price forecast.

We tried a number of specifications for the S&P/Case-Shiller 10-city house price index, which included variables, both considered key determinants of house prices and forecasted in the WEO and simulation models. The specification that we used in this exercise was the following:

In the equation, HP is the 10-city Standard and Poor’s Case-Shiller house price index in log levels, RGDP is real GDP, UR is the unemployment rate, and A2 denotes the second difference.23 Standard errors are reported underneath the coefficient estimates, and ***, **, and * denote statistical significance at the 1, 5, and 10 percent levels, respectively. Under the baseline scenario, using output and unemployment forecasts from WEO, house prices rise moderately but are still 24 percent below their 2006:Q2 peak by end-2014.

The calculations use the observation that developments in house prices generally have fed into CRE prices with a lag when forecasting CRE prices. Consistent with the Global Financial Stability Report’s methodology, the Massachusetts Institute of Technology Center for Real Estate’s Transactions-Based Index of Institutional Commercial Property Investment Performance was used. Since the late 1980s, this price index has closely tracked the Case-Shiller index, with about a four-quarter lag (Table 33.10). Taking this relationship into account, we took actual and forecasted growth rate for the Case-Shiller index and used it to project CRE prices for the same period in the following year. As with house prices, CRE prices increase steadily but are 39 percent off their 2007:Q2 peak and the end of the forecast period.

Table 33.10Correlations between Commercial and Residential Real Estate Prices (Sample: 1987:Q1-2009:Q4)
S&P Case-Shiller House Prices (10-city)Commercial Real Estate Prices
tt + 1t + 2t + 3t + 4t + 5
Levels0.920.940.950.960.960.95
Growth0.430.330.310.410.490.34
Sources: Authors; and Haver Analytics; MIT Center for Real Estate; S&P/Macro Markets, LLC.
References

    Akaike, Hirotugu,1974, “A New Look at the Statistical Model Identification,IEEE Transactions on Automatic Control, Vol. 19, No. 6, pp. 71623.

    Board of Governors of the Federal Reserve System, 2009a, “The Supervisory Capital Assessment Program: Design and Implementation” (Washington, April24). Available via the Internet: http://www.federalreserve.gov/bankinforeg/scap.htm

    Board of Governors of the Federal Reserve System, 2009b, “The Supervisory Capital Assessment Program: Overview of Results” (Washington, May7). Available via the Internet:http://www.federalreserve.gov/bankinforeg/scap.htm

    Breusch, Trevor S.,1979, “Testing for Autocorrelation in Dynamic Linear Models,Australian Economic Papers, Vol. 17, pp. 33455.

    Breusch, Trevor S., and AdrianR. Pagan,1979, “Simple Test for Heteroscedasticity and Random Coefficient Variation,Econometrica, Vol. 47, No. 5, pp. 128794.

    Congressional Oversight Panel, 2010, “February Oversight Report: Commercial Real Estate Losses and the Risk to Financial Stability” (Washington, February). Available via the Internet:http://cop.senate.gov/reports/library/report-021110-cop.cfm

    Durbin, James, and GeoffreyS. Watson,1950, “Testing for Serial Correlation in Least Squares Regression, I,Biometrika, Vol. 37, Nos. 34, pp. 40928.

    Durbin, James, and GeoffreyS. Watson,1951, “Testing for Serial Correlation in Least Squares Regression, II,Biometrika, Vol. 38, Nos. 12, pp. 15979.

    Godfrey, Leslie G.,1978, “Testing against General Autoregressive and Moving Average Error Models When the Regressors Include Lagged Dependent Variables,Econometrica, Vol. 46, No. 6, pp. 1293302.

    Hausman, Jerry A.,1978, “Specification Tests in Econometrics,Econometrica, Vol. 46, No. 6, pp. 125171.

    International Monetary Fund,2008a, Global Financial Stability Report,World Economic and Financial Surveys (Washington, April). Available via the Internet: http://www.imf.org/External/Pubs/FT/GFSR/2008/01/index.htm

    International Monetary Fund,2008b, Global Financial Stability Report,World Economic and Financial Surveys (Washington, October). Available via the Internet: http://www.imf.org/external/pubs/ft/gfsr/2008/02/index.htm

    International Monetary Fund,2009a, Global Financial Stability Report,World Economic and Financial Surveys (Washington, April). Available via the Internet:http://www.imf.org/External/Pubs/FT/GFSR/2009/01/index.htm

    International Monetary Fund,2009b, Global Financial Stability Report,World Economic and Financial Surveys (Washington, October). Available via the Internet:http://www.imf.org/External/Pubs/FT/GFSR/2009/02/index.htm

    International Monetary Fund, 2009c, “United States—Staff Report for the 2009 Article IV Consultation,Country Report 09/187 (Washington, March18). Available via the Internet: http://www.imf.org/external/pubs/cat/longres.aspx?sk=23025

    International Monetary Fund,2010, “United States of America: Stress Testing—Technical Note,IMF Country Report 10/244 (Washington, July).

    Kohn, Donald L.,2010, “Focusing on Bank Interest Rate Risk Exposure,speech at the Federal Deposit Insurance Corporation’s Symposium on Interest Rate Risk Management, Arlington, January29. Available via the Internet:http://www.federalreserve.gov/newsevents/speech/kohn20100129a.htm

    Kumhof, Michael, and DouglasLaxton,2007, “A Party without a Hangover? On the Effects of U.S. Government Deficits,IMF Working Paper 07/202 (Washington: International Monetary Fund). Available via the Internet:http://www.imf.org/external/pubs/cat/longres.aspx?sk=21269

    Lunn, David J., AndrewThomas, NickyBest, and DavidSpiegelhalter,2000, “WinBUGS—A Bayesian Modelling Framework: Concepts, Structure, and Extensibility,Statistics and Computing, Vol. 10, No. 4, pp. 32537.

    Martinez-Peria, MariaSoledad, AndrewPowell, and IvannaVladkova-Hollar,2005, “Banking on Foreigners: The Behavior of International Bank Claims on Latin America, 1985–2000,IMF Staff Papers, Vol. 52, No. 3, pp. 43061.

    Peek, Joe, and EricS. Rosengren,1997, “The International Transmission of Financial Shocks: The Case of Japan,American Economic Review, Vol. 87, No. 4, pp. 495505.

    Schwarz, Gideon E.,1978, “Estimating the Dimension of a Model,Annals of Statistics, Vol. 6, No. 2, pp. 46164.

    Swiston, Andrew,2008, “A U.S. Financial Conditions Index: Putting Credit Where Credit Is Due,” IMF Working Paper 08/161 (Washington: International Monetary Fund). Available via the Internet: http://www.imf.org/external/pubs/cat/longres.aspx?sk=22077

See Board of Governors of the Federal Reserve System (2009a, 2009b).

An earlier version of this framework was used to estimate capital shortfalls in U.S. banks in the context of the Global Financial Stability Report (IMF 2008a, 2008b, 2009a, 2009b) and the 2009 U.S. Article IV Consultation Report (IMF, 2009c). Benefiting from constructive comments from the authorities in the context of the U.S. FSAP, the framework has been revamped to cover a wider range of institutions, better capture institution-specific idiosyncrasies, and account for various regulatory measures and other onetime events during the crisis.

The institutions included in the stress tests all were operating as of end-March 2010. Since then, however, one of the regional banks (W Holding) failed, with its deposits and a portion of its assets acquired by another regional bank (Popular). No adjustment was made for this event, as it would have been difficult to assess its impact, which would depend on the terms and conditions of the takeover.

The asset size of the “small” banks ranges from $10 billion to $60 billion, and market share of assets is as of end-March 2010.

Excluding BHCs with assets smaller than $80 million, which are not reported in the bank-specific database SNL Financial.

We assume that banks would not raise capital over the sample horizon or that, under the baseline, profit-making financial firms would not reduce their dividends policy in anticipation of a future capital need. The latter assumption was relaxed under the adverse scenario, where banks were expected not to pay out common stock dividends, in line with the authorities’ SCAP exercise.

Tier 1 common capital deducts all “noncommon” elements from Tier 1 capital (i.e., qualifying minority interest in consolidated subsidiaries, qualifying trust preferred securities, and qualifying perpetual preferred stock).

We also computed a charge-off rate for “other” loans as a simple average of the other four categories.

The bank-specific fixed effects were obtained from regression analysis.

See IMF (2008a, 2008b) for a description of the methodology used for U.S. securities. This methodology was further revised to better capture losses in non-U.S. countries, particularly Europe and Asia.

A 90 percent confidence interval could yield quarterly return on asset estimates anywhere between 0.25 and 0.75 percent.

See Durbin and Watson (1950, 1951), Godfrey (1978), and Breusch (1979).

Retained earnings, which are defined as the preprovision net revenue minus loan charge-offs, write-downs on securities, taxes, and dividends, can be thought of as the net profits that are returned to capital at the end of each quarter.

For details on the regulatory treatment of DTAs, see Schedules HC-R and HC-F of the FR Y-9C financial statements.

According to U.S. BHC prudential requirements, “allowed DTAs” are to be equal to the lesser of 10 percent of Tier 1 capital (before DTA adjustments) or the amount of DTAs expected to be realized within one year, based on the BHC’s projection of future taxable income.

The possible retrenchment of foreign banks following a shock in the home country is well documented in the literature, including in Peek and Rosengren (1997) regarding the behavior of Japanese banks after the stock market shock in Japan in the early 1990s and in Martinez-Peria, Powell, and Vladkova-Hollar (2005) regarding foreign banks in Latin America.

Kohn (2010) highlighted the potential upward push on interest rates if the rising trajectory of U.S. debt to GDP is not curbed in the future and the impact of higher interest rates on financial intermediaries. Furthermore, the federal banking agencies released policy statements highlighting their expectations for sound practices in managing interest rate risk (http://www.federalreserve.gov/newsevents/press/bcreg/20100107a.htm) and funding and liquidity risk (http://www.federalreserve.gov/newsevents/press/bcreg/20100317a.htm).

This appendix was prepared jointly with Sergei Antoshin (IMF).

The second difference was used after performing an Augmented Dickey-Fuller test on the Case-Shiller 10-city house price inflation rate series, which provided evidence of a unit root.

    Other Resources Citing This Publication