Should the Federal Reserve Reveal More about Its Stress Test Models?
Greg Feldberg, director of research at the Yale Program on Financial Stability, recently testified at a Congressional hearing about a proposal to make the Federal Reserve’s banking stress tests more transparent and put the details out for public comment. In this commentary, adapted from his testimony, he argues that the Fed already discloses more than any other authority in the world about its stress test models and warns that the proposal under discussion could repeat mistakes made in the run-up to the Global Financial Crisis of 2008-09.
Since the Global Financial Crisis of 2008-09, the Federal Reserve has conducted stress tests of the largest banks to evaluate their capital and promote sound risk management, under a mandate Congress created in the Dodd-Frank Act of 2010.
On June 26, a subcommittee of the House Financial Services Committee held a hearing to consider a proposal to put the details of the Fed’s stress scenarios and models out for public comment, in advance, and to disclose significantly more than the Fed currently discloses to the public about its methodologies. The Committee asked YPFS’s Greg Feldberg, a veteran of the Fed and Treasury, to testify on the proposal and the value of stress tests more broadly.
Stress tests have played an important role in our financial history. Along with two of the other panelists who testified last month, I worked on the first stress test as a member of the staff of the Federal Reserve during the Global Financial Crisis. I believe we’re all very proud about how that turned out back in 2009.
I would like to make four points about the state of stress testing.
Supervisory stress tests should be countercyclical. Stress tests should help supervisors keep their guard up during those long periods of financial calm when the possibility of a banking crisis and its associated costs appear remote. In good times, stress tests should be tougher, and we should resist the inevitable pressures to reduce the rigor of the process.
Stress tests should use multiple scenarios. Financial crises tend to come from unexpected places. The Bank of England and European Central Bank have long used “exploratory” scenarios. They are not binding. They are used to help bankers and regulators better understand the risks that are out there. This year, for the first time, the Federal Reserve has also introduced non-binding exploratory scenarios and I welcome that development.
The goal should be to reveal just enough about the Fed’s methods to help banks develop their models and manage their risks, but not so much that it becomes a predictable compliance exercise.
The U.S. innovated the use of supervisory stress tests during the 2008-09 crisis. Early efforts in other jurisdictions didn’t measure up to ours. But in later years the innovation has been elsewhere.
Stress tests are part of a broader regulatory and supervisory toolkit. Some have argued the Federal Reserve’s stress test should have included Silicon Valley Bank, the first large bank to fail last year, and a rising interest-rate scenario. I’m not sure that would have worked. It’s possible that rising rates could have boosted SVB’s net interest income in the test, making up for the decline in asset values.
The International Monetary Fund made that very point about stress tests in 2020 in commenting to the U.S. about its financial stability system. The IMF recommended instead that the U.S. implement an existing Basel standard for interest-rate risk management, with quantitative thresholds, as many other countries have done. I made that argument last year in a blog. There’s no doubt that SVB would have failed the Basel interest-rate risk test.
The lesson is that supervisors need many tools in their toolkit—a lesson that was driven home recently when supervisors rejected several big banks’ resolution plans.
Transparency can be a double-edged sword. As I noted in a paper in 2019, the goal should be to reveal just enough about the Fed’s methods to help banks develop their models and manage their risks, but not so much that it becomes a predictable compliance exercise.
I’ll make two observations about that tradeoff. First, we are already revealing a lot to the regulated industry, which may allow them to merely optimize to the test. If you read the Fed’s public methodology and scenario documents, you’ll see 50 pages describing the models, including descriptions of inputs and equations, and 26 pages on loss rates. I had a colleague on my team look at other jurisdictions to see what they revealed. He found that the U.S. was far ahead of every other country, with the exception of the Bank of England to some extent, and he couldn’t find another country that revealed its loss rates.
So you might wonder what the witnesses in last month’s hearing would like the Fed to be more transparent about. The written testimony of one witness was very helpful. He wrote, in essence, we do get a lot of information about models and loss rates, but we’re missing the parameters. Just tell us the parameters, he urged, and we’ll really be able to follow Grandma’s secret cake recipe.
I believe that would be a mistake. It simply wouldn’t be good policy to reveal so much about the inputs in a supervisory stress test. At some point it no longer has any impact.
We have a good example of what happens when you put the parameters of a stress test out for public comment. The failed government-sponsored enterprises Fannie Mae and Freddie Mac were subject to Congressionally mandated stress tests from 2002 to 2008. But these stress tests did not detect the growing risks the GSEs were taking. In fact, the tests showed strong capital compliance.
Why? It turns out the supervisor, the Office of Federal Housing Enterprise Oversight, never tweaked its models, despite growing, obvious risks in the mortgage market. One paper found that the requirement to “publish all stress scenarios, empirical specifications, and parameter estimates in the Federal Register” meant that “it would have been administratively cumbersome to make any material changes to the underlying models.”
In short, supervisory stress tests are very important for bank risk management. I’m concerned, as are many others, that the tests have become too routinized and that further disclosures from regulators will simply make them more so.