Notes from the Vault
Larry D. Wall
December 2013
Although some observers have argued that simpler measures would be better, the United States is now committed to using two relatively complex and costly approaches to measuring capital adequacy: Basel III and stress tests. How did we come to have two measures? Can stress testing, the more flexible of the two measures, be used in a way that adds incremental value to Basel III?
Two of my recent working papers address these issues. "The Adoption of Stress Testing: Why the Basel Capital Measures Were Not Enough" explains why the United States turned to stress tests rather than the Basel capital ratios to address market concerns in 2008–09 that U.S. banks were undercapitalized. "Measuring Capital Adequacy Supervisory Stress Tests in a Basel World" identifies several problems with the Basel III approach and critically evaluates the ability of stress tests to address these problems. This month's Notes from the Vault summarizes some of the key findings in those papers.
Two different approaches
Basel III and stress testing as implemented in the United States both rely on projections of losses in an extreme scenario to evaluate the adequacy of an individual bank's capital. Both measures require the estimation of statistical models. However, the two measures differ in some fundamental ways.
Basel III provides three different measures of capital adequacy, all of which are based on the historical loss distributions. The standardized approach relies exclusively on risk weightings derived from supervisory judgments about the relative riskiness of different bank exposures. The internal ratings based (IRB) and advanced IRB require banks to use historical data to estimate the level of losses on an asset that should be exceeded only once every 1,000 years. These extreme loss estimates are then used to derive risk weights that are applied to each asset (or more generally, each exposure). The risk-weighted assets are then summed to calculate the denominator of the Basel III ratios. In effect, Basel III derives a generic severely adverse scenario for each portfolio category from that category's own (recent) past experience.
In contrast, the calculation of the various measures of capital used in Basel III is a mechanical calculation. Each of the capital measures calculates capital as the sum of certain balance sheet items (such as common equity and retained earnings) minus certain other items (such as certain intangible assets).
The stress tests begin with several different regulatory measures of capital adequacy, including a Basel III measure in recent tests. However, rather than relying exclusively on current balance sheet information, the stress tests project future book capital under each of several different scenarios. The estimates of the change in accounting capital are based in part on estimates of each bank's losses in each portfolio given a particular scenario. However, given that the stress tests are forward looking, they also include estimates of each bank's likely earnings, called their pre-provision net revenue (PPNR).
This brief summary suggests that stress tests may add incremental value in two situations. First, accounting measures of capital may deviate significantly from economic values. Basel III relies entirely on accounting values, whereas the stress tests can reveal unrecognized losses by measuring losses that would be recognized in the future.
Second, although supervisors should be concerned about the full range of risks facing banks, they may at times have good reasons to be especially concerned about a subset of risks. Basel III's risk measures provide information about a broad range of risks while the current stress tests can provide information about subsets of risk. One way of considering this is to think of the two tests as casting light on banks' risk exposures. Basel III casts a dim light over a wide range of possible scenarios. Basel III's use of historical loss distribution data allows it to estimate a lower bound on once-in-1,000-year losses across a wide variety of scenarios. However, Basel III cannot say anything about what may happen in any particular scenario. In contrast, each individual stress test casts a very bright light, but only on one particular scenario. Scenarios similar to the one being tested are likely to produce similar results, but that outcome is not guaranteed. Still, there is no reason to expect that any given scenario will be predictive of the results of a very different stress scenario.
Using stress tests to identify losses
Many large banks experienced significant liquidity problems in late 2008 as investors became concerned about their financial condition, despite the fact that most banks continued to report capital ratios well above supervisory minimums. One could argue that the problem was that the United States had not yet adopted Basel II, therefore, it was underestimating the riskiness of banks assets. The problem with this claim is that U.S. banks generally showed higher capital ratios under Basel II than Basel I. Rather, the problem was that many investors were concerned that bank asset values were overstated, which implied that their capital ratios were also overstated.
The 2009 concerns about overstated asset values were a problem that could be addressed by a stress test, but not by any version of the Basel ratios. The key to identifying overstated asset values is for the stress scenario to last sufficiently long and be sufficiently adverse so that current problem assets cannot recover in value. The Supervisory Capital Assessment Program (SCAP) in 2009 set out a two-year scenario, explicitly for the purpose of capturing "a large portion of losses from positions held as of the end of 2008" (page 3). It also specified a sufficiently adverse scenario such that asset values were unlikely to recover. The resulting estimates appear to have had some credibility in financial markets and helped restore confidence in U.S. banks.
Although stress testing was helpful in identifying as yet unrecognized losses in 2009, the tests are unlikely to make as big a contribution in a future crisis. In part, this is because the Financial Accounting Standards Board is likely to change the accounting rules to require earlier loss recognition, as discussed in a previous Notes from the Vault. Another reason is that the SCAP took place in almost ideal conditions for the supervisors to run a stress test designed to identify unrecognized losses. Conditions were ideal in that the stress test was unlikely substantially to reduce market confidence in banks from its depressed level in early 2009, and the supervisors had a credible mechanism for dealing with capital deficiencies revealed by the stress tests in the form of the Capital Purchase Program of the Troubled Asset Relief Program (TARP). If either or both of these conditions are not met next time, supervisors may be less likely to test scenarios that could reveal large, unrecognized losses.
Using stress tests to mitigate moral hazard
The credit risk measures in Basel III are likely to contain both random errors and estimates that may be biased down. Stress tests differ from Basel III in a variety of ways that could mitigate the adverse effects of those cases where Basel III underestimates credit risk.
The process of obtaining Basel III risk weights is likely to result in a large number of random errors. The standardized approach of Basel III is a one-size-fits-all model based on supervisory judgment that is unlikely to match perfectly the riskiness of all the individual portfolio items in any bank at any given point in time. The IRB approaches are intended to be more sensitive to each bank's risk exposure, but the Bank of England's Andrew Haldane estimates that some banks could have over 200,000 Basel III risk buckets. Inevitably, estimating all of the parameters needed for these buckets will give rise to both over- and underestimates of risk. Stress test loss estimates will also contain random errors, but these errors are unlikely to be perfectly correlated with the Basel III errors. As a result, stress tests can mitigate some of the bad incentives created by Basel III random errors that understate risk.
An analogous problem with the Basel III IRB risk weights specifically is that they are estimated by banks that have an incentive to produce lower risk estimates. To be sure, banks must obtain supervisory approval for their models, but supervisors cannot review all of the choices made by a bank in developing its Basel III models. The stress tests also rely in part on estimates supplied by the banks, as well as on estimates from the Federal Reserve's independent models. Additionally, the supervisory stress tests are conducted at all large banks at the same time, which facilitates supervisory comparisons across banks. As a result, the stress tests put supervisors in a better position to identify banks that appear to be lowballing their risk estimates. Moreover, the stress tests are only a part of the Comprehensive Capital Analysis and Review (CCAR). A bank that passes the stress test portion may nevertheless fail the overall CCAR if its stress models are found to be unsatisfactory.
Using stress tests to go beyond recent historical experience
The risk estimates from Basel III are based on relatively recent experience, in part because in many cases banks and supervisors have data covering only the last two decades. The data do not include some potentially relevant high loss scenarios. For example, prior to the recent crisis, the United States had not experienced a nationwide downturn in housing prices since the 1930s. The limited data on credit history is a problem that may potentially be mitigated with stress tests using a scenario that has not been observed in the recent data. Of course, data limitations may also affect the parameters in the stress test model, for example, by making it difficult to identify nonlinear relationships between the stress scenario parameters and credit losses. Supervisors may be able partially to compensate for parameters that likely underestimate risk by making the scenario even more adverse (such as specifying a 20 percent housing price decline to obtain estimates of the likely losses from a 10 percent decline).
Another supervisory concern that lies outside recent experience is that of a large increase in the level and slope of the term structure. A reasonable concern is that banks have become vulnerable to such an increase, given that the United States is entering the sixth year of a monetary policy regime at the zero lower bound. Basel III currently lacks a charge for interest rate risk. While in practice the stress tests are not well designed to measure the effect of changes in interest rates, in theory they could be.1 Currently, supervisors have other tools for addressing interest rate risk. However, these supervisory tools do not provide the same automatic link to a bank's capital requirements that exists with Basel III and the supervisory stress tests.
Concluding comments
The existing stress tests have the potential to mitigate weaknesses in Basel III's measurement of capital and credit risk. A modified version of the stress tests may also be used to bring banks' interest rate risk into the formal capital requirements. However, a realistic assessment of the stress tests' role in mitigating Basel III weaknesses should also include a couple of cautionary notes. First, banks can adjust their portfolios to reduce their exposure to risks well measured by the stress tests but increase their exposure to risks poorly captured by the current version of the stress tests. This could take the form of increasing exposure to assets whose risk is likely underestimated by the stress models; albeit supervisors can partially compensate for this by imposing a more stringent scenario as discussed above. Additionally, banks could increase their exposure to possible economic shocks that have not been included in recent scenarios.
Second, the success of the stress tests depends upon their design and the specific scenarios being tested. The tests can also easily be designed to avoid revealing weaknesses in banks' capital adequacy. All that the supervisors need to do is structure the tests so that they do not measure losses in the weaker parts of the portfolio or design a "stress" scenario that fails to stress the weaker parts. For example, the European Union conducted stress tests in 2010 and 2011 that "significantly underestimate problems in the financial sector," according to Moody's Enam Ahmed and coauthors, only to be followed within months by the failure of banks that had passed the EU's most recent stress test.
Larry D. Wall is the executive director of the Center for Financial Innovation and Stability at the Atlanta Fed. The author thanks Mark Jensen and Paula Tkac for helpful comments on the paper. The views expressed here are the author's and not necessarily those of the Federal Reserve Bank of Atlanta or the Federal Reserve System. If you wish to comment on this post, please e-mail atl.nftv.mailbox@atl.frb.org.
_______________________________________
1 Part of the problem is that the stress tests rely on accounting values. As a result, the stress tests can measure only that part of the loss that would be recognized in accounting statements during the stress horizon.