Stress testsIf you are a risk manager at a big bank, you might want to run some stress tests examining how your bank will perform under stressful scenarios. If you are creative and good at your job, you will have fun thinking up the scenarios. You will get in a room with some of your subordinates, and ideally with some front-line traders and managers, and brainstorm bad stuff that might happen.
Your brainstorming might include some history: “What if 2008 happens again?” “What if 1929 happens again?” “What if LTCM happens again?” It might include some simple numerical questions: “What if the S&P falls by 20%?” “What if the Fed raises rates to 7%?” “What if the Fed lowers rates to 0%?” You might think about social and geopolitical and technological questions and try to translate them into economic scenarios: “What if nobody goes back to the office and office rents fall
50%?” “What if artificial intelligence causes mass unemployment?” “What if Russia’s war in Ukraine keeps pushing up oil prices?” You might think about scenarios specific to your bank’s operations: “What if our CEO gets run over by a bus?” “What if our CEO steals the corporate treasury?”
Have fun, go nuts, be creative. Think of lots of scenarios. Then model how those scenarios will translate into market prices, how they will affect your funding costs and the cash flows from your assets. Then model how much capital and liquidity you will have in each scenario. Part of the goal here is to make sure that you will have plenty of capital and liquidity in a wide range of stressful scenarios. Part of the goal is to figure out which scenarios will be worse for you, so you can know what to worry about and hedge: If your model tells you that you’ll be fine if the Fed raises rates and bankrupt if it lowers rates, maybe you should do something about that.
If there is some specific event that you are worried about — some new worry that crops up — you might sit down and design a stress test for that event to make sure that you’d survive it. But your overall approach to stress testing will be something like “let’s constantly think of new things to worry about, and test for those.” It will not be “let me think of the one biggest thing to worry about, and test only for that.” Lots of things can go wrong in different directions! If you only worry about one thing, you will miss the other things.
Also separately the Federal Reserve conducts annual stress tests for big US banks, but those are … different. The Fed’s stress tests were created after the 2008 crisis, basically to shore up confidence in the banks so that they could raise capital. The Fed is in the business of supervising big banks, and as part of that business it prods the banks to consider various risks, to prepare for different scenarios, to build a robust culture of stress testing and risk management. But the Fed’s official stress tests are a public exercise designed to make sure — and tell everyone — that the banking system could survive another 2008.
And so each year the Fed sits down and thinks something like “what is the most plausible way for 2008 to happen this year,” and then it writes one stress-test scenario[1] that is basically “there’s a recession and real estate prices collapse,” and the banks run their models to see how much money they would lose in that scenario, and generally the answer is “a certain amount, but not enough to leave us undercapitalized,” and the banks pass the stress tests.
This became slightly awkward in 2020 when the stress-test scenario involved the US unemployment rate peaking at 10%, while the actual stress of Covid-19 caused unemployment to peak at 14.7%. If you ask questions like “can banks survive another 2008,” you will not quite get an answer to the question “can banks survive a global pandemic?” The answer to that question, however, turned out to be yes: For a combination of reasons (supportive government policy and Fed lending, but also profiting on volatility), the big banks did pretty well in Covid.
It also became slightly awkward this year when the Fed released its stress-test scenario in February, and it went like this:
The severely adverse scenario is characterized by a severe global recession accompanied by a period of heightened stress in both commercial and residential real estate markets, as well as in corporate debt markets. The U.S. unemployment rate rises nearly 6½ percentage points from the starting point of the scenario in the fourth quarter of 2022 to its peak of 10 percent in the third quarter of 2024. The sharp decline in economic activity is also accompanied by an increase in market volatility, widening corporate bond spreads, and a collapse in asset prices, including a 38 percent decline in house prices and a 40 percent decline in commercial real estate prices. …
The rising unemployment rate and the rapid decline in aggregate demand for goods and services significantly reduce inflationary pressures. …
Short-term interest rates, as measured by the 3-month Treasury rate, fall significantly to near zero by the third quarter of 2023 and remain there for the remainder of the scenario. Long-term interest rates, as measured by the 10-year Treasury yield, fall by nearly 3¼ percentage points by the second quarter of 2023, and then gradually rise in late 2023 to about 1½ percent by the end of the scenario.
Fine, right, recession, real-estate collapse, corporate credit blows out, rates go to zero, the 2008 of 2023. But then in March a couple of US banks failed, and many others came under a lot of stress, for exactly the opposite reason: Inflation was persistently high, interest rates went up, and those banks had to pay more on deposits even as their holdings of Treasury bonds lost value. Silicon Valley Bank was not subject to the Fed’s stress tests, as a mid-sized bank, but if it had been it would have done great! It did not do great in real life.
Anyway yesterday the 23 big banks subject to the stress tests all passed. Here are the Fed’s press release and the full results. One of the banks that passed is the US arm of Credit Suisse Group AG, which survived the Fed’s hypothetical stress scenario just fine but which vanished due to the stress of real life. From the Fed’s press release:
"Today's results confirm that the banking system remains strong and resilient," Vice Chair for Supervision Michael S. Barr said. "At the same time, this stress test is only one way to measure that strength. We should remain humble about how risks can arise and continue our work to ensure that banks are resilient to a range of economic scenarios, market shocks, and other stresses."
Yes.
Now, this year’s stress test did include another scenario for the eight biggest banks, which (1) is different from the regular stress test and (2) is closer to what actually happened:
In February 2023, for the first time, the Federal Reserve published an additional, exploratory market shock component that applied only to U.S. G-SIBs. The purpose of the stress test is to understand a firm’s resilience to a range of severe but plausible events, and the exploratory component furthers that purpose by posing a different set of risks than is probed by this year’s global market shock component.
In contrast to this year’s global market shock component, which was characterized by a severe recession with fading inflation expectations, the exploratory market shock is characterized by a less severe recession with greater inflationary pressures induced by higher inflation expectations, increases in interest rates, an appreciation of the U.S. dollar, and increases in commodity prices.
This scenario doesn’t count for bank capital requirements — “Consistent with the nature of an exploratory exercise, the exploratory market shock will not contribute to the capital requirements set by this year’s stress test” — but it is intellectually interesting. Broadly speaking the big banks did a little better in this scenario than in the severe-recession scenario.
But also, in real life, the big banks did fine in the scenario that actually happened (except Credit Suisse!): The problems at US regional banks seem to have driven deposits into the biggest banks, as depositors got nervous about small banks and fled to the safety of big banks. (Look how safe they are, they keep passing the stress tests.) The actual world of 2023 was different from the hypothetical world of the stress tests, and in some ways more stressful for the US banking system, but it worked out fine for the biggest banks.
A general feature of the Fed’s stress tests is that, when the Fed serves up a “severely adverse” stress scenario to the banks, the banks go off and model the effect it would have on their business, and if they come back to the Fed and say “actually this would be good for us,” the Fed gets mad at them. You’re not supposed to say that! Goldman Sachs Group notably did this in 2020, arguing that its trading business was countercyclical, and that in a stressful scenario it would simply make a ton of money trading derivatives; the Fed was not amused. But sometimes it is true! Some hypothetical scenarios would be very bad for the big banks, other hypothetical scenarios would be good for the big banks, and it’s perfectly plausible that some scenarios would be bad — for the world, for the economy, for the banking system — while also being good for the big banks. The way to manage this risk is to think about lots of scenarios and make sure the banks would survive all of them.
