The media printed the headline, summarized the data, and moved on when the Federal Reserve released the results of its most recent round of stress tests. All of the major American banks passed. Under “severe” hypothetical shocks, capital ratios remained stable. On paper, the system absorbed everything, including a recession, a property crash, and an increase in unemployment. If you were to read the briefing notes, you would believe that the American banking industry had become nearly unbreakable.
It hasn’t. And those who create these tests are aware of it.
You begin to notice something when you speak with economists who work in this field. They are more direct in private and more circumspect in public. For the past few years, there has been a growing perception that stress tests have shifted from being diagnostic instruments to something more akin to theater. In a recent article for NIESR, Dr. David Aikman stated out loud that the results communication process has “the whiff of a PR exercise.” It’s a powerful statement from the macroprudential realm, where being modest is the norm.
The figures speak for themselves. On paper, scenarios have become more severe—seven standard deviation shocks, simultaneous market crashes, global GDP collapses—but banks continue to pass them with plenty of leeway. Aikman cites the Bank of England’s 2021 exercise, in which a historically severe shock only slightly reduced capital. At the same time, market-based estimates indicated a fifty to sixty billion pound capital shortfall. Two images, identical banks, but quite different conclusions.
There is an uncomfortable fact about the American case that is readily apparent. Silicon Valley Bank. According to Tomasz Piskorski of Columbia, the Fed’s own stress framework would have most likely deemed its balance sheet “fine” three months prior to its collapse in 2023. According to the standards regulators were using, the bank that ran out of money during a routine interest-rate adjustment was completely safe. That particular detail ought to have prompted a fundamental reconsideration. Rather, this year’s tests continue to concentrate solely on the thirty-one banks with assets exceeding $100 billion, virtually completely excluding the regional lenders with the greatest exposure to commercial real estate—the very risk that everyone has been secretly concerned about since 2023.
It’s difficult to ignore the pattern. The largest banks are put to the test in scenarios that are shaped by their economists. They get through. The outcomes are made public. Investors unwind. The Financial Policy Committee in the UK did just that in December, reducing a system-wide buffer in part due to encouraging stress-test results. Regulators use these same results as justification for relaxing other regulations. It is uncomfortable to watch the feedback loop.
What the tests presume people will do in a crisis is another issue. The majority of models assume banks will cut dividends and bonuses without hesitation and spend down their capital buffers smoothly. Bank executives don’t act that way during actual downturns. They hoard, cut back, and safeguard their own income and stock prices. This was amply illustrated by the 2008 crisis, but the behavioral presumptions found in current models are essentially the same.
The aspect of the system that isn’t being tested at all gets lost in the optimism. lenders who are not banks. funds for private credit. concentrated real estate loans from regional banks. Stress in nonbank financial institutions has already begun to affect bank stock prices, according to a recent observation by the Federal Reserve Bank of New York. These connections exist even though the stress tests do not account for them.
As this develops, the concern isn’t so much about the strength of the major banks. In the limited sense that the tests gauge, they most likely are. The larger concern is that a tool designed to reveal hidden risk has gradually evolved into a tool for verifying what markets and regulators already want to believe. That’s how complacency usually develops. Not by ignorance. by providing assurance.
