The Fed’s bank stress test proposal would only heighten uncertainty

The justifications for the proposed changes to large bank regulatory capital requirements in the so-called “Basel III end-game” Notice of Proposed Rulemaking are schizophrenic. 

Regulators argue that bank internal model credit risk capital requirements must include a new standardized approach to determine a floor for the credit risk capital a bank must hold. The change is allegedly needed because banks’ internal credit model loss estimates are inaccurate and virtually impossible to verify. 

Fair enough. Yet the same notice ignores these concerns when it expands the use of the Fed’s own secret internal stress test loss models to decide whether large bank stress capital buffers should be higher than 2.5 percent.

The process of determining bank stress capital buffers lacks transparency because the Fed’s stress test loss models are secret. The transparency of this process will take on added importance when the Fed adds new climate change scenarios into the annual stress test exercises it uses to set banks’ stress capital buffer requirements.  

The justification for replacing internal credit risk models with a standardized approach for setting credit risk capital requirements also applies when evaluating the use of model-based loss estimates to calibrate stress capital buffers. The magnitude of the uncertainty attached to stress model loss estimates — be they bank internal models or the Federal Reserve Board’s own models — is massively greater than the uncertainty attached to loss estimates from internal credit risk models.

Bank internal credit risk models estimate credit losses using actual bank loss historical data. The estimation uncertainty in stress test loss estimates is magnified because stress model losses are, by definition, estimated over very long time horizons assuming extreme economic conditions that have never happened. Consequently, there are no benchmark observations available to validate stress loss estimates. 

Truth be told, even if David Copperfield were the Fed’s vice chair of supervision, the Fed would still be unable to verify the accuracy of its estimates of future bank losses under conditions so severe they have never before happened.  

If model-based loss estimates are so inaccurate and impossibly difficult to verify that regulators cannot trust them to set minimum regulatory capital requirements, there is no logical justification for using them to raise large banks’ stress capital buffers using secret Federal Reserve Board internal models if transparency and accuracy are policy goals. Consider how the Fed describes its own internal stress test models:

• “[These models do] not place undue emphasis on historical outcomes in predicting future outcomes. 
•  The Federal Reserve aims to produce supervisory stress test results that reflect likely outcomes under the supervisory scenarios…. that have not occurred.”

The Fed publicly states that its stress test models are not designed to accurately reproduce actual historical bank losses experienced under actual historic economic conditions but claims that these models accurately estimate future bank losses for nine consecutive quarters under economic conditions so dire that they have literally never before been observed. Unless you are a consultant or a Fed economist paid to indulge in these fictional modeling exercises, it is hard to swallow such claims.      

Consider the Notice of Proposed Rulemaking text (p. 16) that justifies the adoption of a standardized credit risk approach. The notice’s test itself makes my case if you replace the italicized words below with “stress test”:

• “[I]nternal [Stress test] models rely on a … choice of modeling assumptions and supporting data. Such model assumptions include a degree of subjectivity … empirical verification of modeling choices can require many years of historical experience because severe credit risk and operational risk [stress test] losses can occur infrequently.”
• “The complexity of these [stress test] models-based approaches, can reduce confidence in the validity of … [and] lessen the transparency of the risk-based capital ratios, and challenge comparisons of capital adequacy across banking organizations.”
• “The use of robust, risk-sensitive standardized approaches for credit and operational risk [stress tests] would also improve the efficiency of the capital framework by reducing operational costs.”
• “Replacing the use of internal [stress test] models with standardized approaches would reduce costs associated with maintaining such modeling systems.”  

If adopting a standardized approach is necessary to set an accurate transparent floor under internal model credit risk capital requirements because internal model estimates are not to be trusted, then common sense suggests that it is inappropriate to deputize the Federal Reserve Board with the power to increase the stress capital buffers of the largest banks using stress scenario loss estimates from secret and unverifiable internal models. 

The Fed should adopt and fully disclose a standardized method for determining whether a bank requires a stress capital buffer larger than the 2.5 percent minimum. This transparency issue will take on elevated importance once the Fed introduces new climate-change scenarios into its annual stress tests. Without full transparency, the Fed will be unaccountable should it covertly use the stress capital buffers to raise minimum capital requirements on banks that dissent from the administration’s climate-change agenda.

Paul Kupiec is a senior fellow at the American Enterprise Institute specializing in financial services issues.