In February, the Federal Reserve published the macroeconomic scenarios for the 2023 stress test of large banks under the Dodd-Frank Act. In the “severely adverse” scenario, there is a steep decline in interest rates, which return to low levels for the full three years of the scenario. In the immediate aftermath of the collapse of Silicon Valley Bank (SVB), some commentators noted wryly the mismatch between this 2023 stress and the actual sharp rise in interest rates over the past couple of years, which was a major factor in bringing SVB down.
For reasons that need not detain us here, banks of SVB’s size were not included in these stress tests. But the question remains: Just how appropriate is the choice of macroeconomic scenarios?
But it’s not only for 2023 that this feature appears. Indeed, every severely adverse scenario used by the Fed since 2015 has the 3-month Treasury bill rate ending up at 0.1 percent. Many historic episodes of severe economic downturn have indeed been accompanied by low interest rates, as the Fed used its policy tools to support aggregate demand. But it is a bit strange that not since 2015 has a stress test involved rising interest rates.
One of the advantages of stress testing the banks every year is that their robustness to a variety of contrasting stresses can be assessed. Just repeating a broadly similar scenario year after year misses the opportunity to provide supervisors with potentially important information on vulnerabilities. It can also result in policymakers assuming that the banks are robust to more types of shock than is really the case.
Yet pointlessly repeating a broadly similar scenario each year is exactly what the Fed has been doing, as we can show here.
It’s not straightforward to make this comparison, as each year the scenario has 28 variables, each modelled for 12 quarters. In order to get a visual representation of the main features of each scenario, I have selected five key variables from the 28: GDP growth, unemployment, short-term interest rate, stock market, and housing prices. To compress the timelines, I choose the cumulative growth rate in the first four quarters for GDP, the levels after four quarters of the interest rate and equity prices, and the levels after eight quarters of unemployment and housing prices. (The choice of four and eight quarters is intended to capture the typical length of time for the variable to reach its trough.)
With these choices, I compare each variable in the severely adverse scenario to its value in the baseline (difference for GDP growth, interest, and unemployment rates, percentage difference for equity and housing prices). Finally the results are expressed as a proportion of its average over the nine stress tests from 2016 to 2023 and plotted on this radar chart.
The result shows how similar the pattern of macroeconomic stresses chosen by the Fed has been from year to year. If we ignore interest rates for a moment, we see almost no change in the pattern from year to year. The stresses were a bit smaller back in 2016-19, than they have been since, but the pattern, i.e., the degree of stress in the four main variables, is about the same.
Only in interest rates is the pattern of stress different from year to year. But even here, the differences have a common cause, namely that, no matter where it begins, the 3-month Treasury bill rate is assumed every year in the severely adverse case to fall to just above zero. It’s effectively the same stress.
To be sure, there are some other differences when all of the 28 variables in the macroeconomic scenarios are examined, and there are some moderate variations over the 12-quarter duration of the stress scenarios. But for the main picture, the banks are being tested on fundamentally similar types of shock every year. No wonder the vulnerability of the business model of some banks has flown under the supervisory system’s radar.