By Warren Hrung and J.P. Rothenberg
The turmoil in the financial sector driven by a few large bank failures has raised broader questions about the current supervisory stress testing framework. Silicon Valley Bank was one of the largest bank failures in U.S. history and was not subject to an annual supervisory stress test. In Switzerland, Credit Suisse, a global systemically important bank, was purchased by UBS in a “forced marriage” after a loss in investor confidence. These events provide a valuable opportunity to compare the frameworks for bank stress testing in the U.S. and Switzerland and identify aspects that should be examined in any regulatory reform proposals.
A stale starting point for U.S. stress tests
Recent commentary regarding the collapse of SVB has noted that subjecting SVB to the annual bank stress testing exercise would still not have highlighted the bank’s risk of failure. This is because the 2022 scenarios — the last stress testing exercise to be completed before the bank’s collapse — did not include the abrupt increase in interest rates that ultimately led to SVB’s demise.
It is worth digging deeper into why the stress scenarios, which focus on bank capital adequacy and not liquidity issues, did not include such an extreme move in interest rates. The current systemic stress has been building in the banking sector since the Federal Open Market Committee began increasing interest rates in early 2022. The stage was arguably set even earlier as the FOMC targeted a range of 0-0.25 percent for the federal funds rate for much of the post-2008-09 financial crisis period. During this period, the maximum target range for the federal funds rate only reached 2.25-2.5 percent before falling back to the zero-lower bound during the COVID-19 pandemic. Starting in early 2022, the FOMC rapidly raised the policy rate 500 basis points by early May 2023 in response to the persistent inflation that the Federal Reserve had initially thought to be “transitory.”
The key to understanding why the supervisory stress scenarios did not include a dramatic increase in interest rates can be found in the underlying framework that governs the design of the stress test scenarios. This framework states that the general approach is to reflect the post-war U.S. experience with recessions and that the unemployment rate is “the primary basis for specifying the severely adverse scenario.” This is because an increasing unemployment rate is a typical feature of recessions. After the path of the unemployment rate is specified, the paths of the other macroeconomic variables in the scenario are calibrated to be consistent with the path of unemployment.
Note that a scenario within the stress testing framework cannot simply assume extreme paths for all macroeconomic variables. For example, an increase in the unemployment rate and interest rates to 50 percent, and a 90 percent fall in the U.S. dollar and equities would be a difficult scenario to justify as plausible absent, say, a zombie apocalypse — at which point bank conditions would be the least of our worries. The assumptions for the stress scenarios must take into account the economic conditions at the time the scenarios are designed and must exhibit internal consistency, even if unprecedented movements and levels are assumed.
To illustrate why it is unlikely that an extreme increase in unemployment would be associated with a dramatic increase in interest rates, we have calculated that the correlation between the changes in the unemployment rate and changes in the three-month Treasury rate going back to 1948 is slightly negative (-0.10). So, an increase in the unemployment rate is typically associated with a decrease in the three-month Treasury rate. This calculation includes the post-financial crisis period where the three-month Treasury rate was often close to the zero-lower bound. Removing the post-2008 period doubles the correlation coefficient (-0.20). Thus, any coherent and “plausible” narrative that is premised on an increase in the unemployment rate would be unlikely to also contain an increase in interest rates, much less a dramatic increase in interest rates.
If future stress test exercises are to involve scenarios that are more likely to uncover hidden risks in bank portfolios, supervisors should consider varying the macroeconomic variable that serves as the starting point of the scenarios. For example, instead of starting with an increase in the unemployment rate, the scenarios could be based on changes in other macroeconomic variables, such as inflation or shifts in the Treasury yields, that can stress bank portfolios in different ways. The precedent for doing so is arguably available, based on the experiences of the late 1970s and early 1980s, when interest rate movements were driven more by inflation and rose dramatically while the unemployment rate was falling.
However, supervisors should also work to ensure that any change in approach is used to inform bank management about potential risks and does not result in excess volatility in capital requirements from year to year. Multiple stress scenarios would provide a possible vehicle for varying the focus of stress while also limiting the volatility in loss estimates for a given bank. In the Federal Reserve’s retrospective report, Vice Chairman for Supervision Michael Barr mentioned that the Federal Reserve was already looking at the use of multiple scenarios in stress testing, but he did not provide further details.
A lack of transparency in the Swiss framework
In Switzerland, the Swiss Financial Market Supervisory Authority, or FINMA, is the primary regulator for the financial system and is responsible for the supervision and regulation of banks. FINMA collaborates with the Swiss National Bank, which sets policy interest rates and has a financial stability mandate. This system oversees banking assets that were in excess of 500 percent of Swiss GDP in 2020. For comparison, U.S. banking assets are just over 100 percent of US GDP.
While banking assets are a relatively high percentage of GDP, Switzerland’s regulatory authorities only have about four full-time employee equivalents overseeing the two domestic stress tests, compared to more than 40 FTEs for the U.S. and over 300 for the European Union area stress test, according to the Bank for International Settlements.
The SNB started systematically stress testing banks in 2008. Currently, both FINMA and the SNB have stress testing programs to assess banks’ resilience to adverse market conditions that help evaluate banks’ capital and liquidity positions. The Swiss financial authorities typically provide the banks with the stress test scenarios and share the results with each other. The SNB runs two stress testing programs referred to as the “Building Block Analysis,” which are supervisor-run exercises where the SNB defines and runs the scenarios using SNB models with the bank’s reported data. One BBA focuses on the Swiss G-SIBs (UBS and, until this spring, Credit Suisse), while the other focuses on the non-G-SIB Swiss banks. FINMA runs its own stress testing program that focuses on the two Swiss G-SIBs using an institution-run approach where FINMA defines the scenarios but relies on the banks to report the results using internal data and models. Of note is that liquidity testing is not a primary consideration in these exercises but is part of other monitoring programs.
Credit Suisse and UBS were subject to several stress testing regimes. Both had U.S. subsidiaries that complied with CCAR/DFAST and which they had recently passed. Both banks were also subject to the domestic BBA and the FINMA stress test, according to a BIS comparative analysis and presumably passed. Further, the IMF conducted a number of Financial Sector Assessment Program reports, with the most recent being published in 2019 and mentioning first-time access to confidential bank data.
Information about Switzerland’s stress testing programs is not as readily available as in jurisdictions like the U.S. or EU. While the SNB has been publishing financial stability reports since at least 2003, FINMA does not publish the scenarios or the results of the UBS or Credit Suisse stress tests. Just prior to the collapse of Credit Suisse, FINMA issued a statement that the bank “meets the capital and liquidity requirements imposed on systemically important banks.” Without more transparency about the stress testing framework in Switzerland, it is impossible for the public to study and assess them.
Conclusion
In the U.S., developing stress testing scenarios that start with stresses to other macroeconomic variables instead of the unemployment rate can help reveal risks that would be difficult to uncover in the current framework. Meanwhile, in Switzerland, increased transparency regarding the stress testing framework and results is an important first step towards identifying other aspects of the stress testing exercise that could be enhanced.
Warren Hrung is SVP and head of banking and financial services research at ABA. J.P. Rothenberg is VP for economic research at ABA.
