Stress Tests Ask What Breaks in a Scenario Somebody Wrote Down
Regulators define a hypothetical severe downturn and require banks to show they would survive it. The exercise is valuable and it can only test the scenario that was imagined.
What the Exercise Does
A supervisor defines a hypothetical scenario, typically several years of severe economic deterioration: unemployment rising sharply, asset prices falling, a market shock.
Banks then project their revenues, losses, and capital positions under that scenario, and must demonstrate that capital remains above the required minimum throughout.
The results determine whether a bank may pay dividends and repurchase shares, which is what gives the exercise force.
The binding consequence is on distributions, not on lending. That is what makes stress tests something banks manage toward rather than treat as a modelling exercise.
Why They Started
The first large scale exercise of this kind was conducted in the United States in 2009, when the central problem was that nobody knew which banks were sound.
Publishing the results, including individual institution capital shortfalls, resolved that uncertainty. Banks found short of capital were required to raise it, and the disclosure itself restored a degree of confidence because the numbers were credible.
The subsequent institutionalisation of the exercise made it a recurring supervisory tool rather than a crisis measure.
What They Cover
| Component | Tested |
|---|---|
| Credit losses | Loan portfolio under stressed conditions |
| Trading losses | Market shock applied to positions |
| Counterparty default | Failure of the largest counterparty |
| Revenue under stress | Net interest income and fees |
| Operational risk losses | Modelled, historically significant |
The Structural Limitation
The scenario is written in advance by people who cannot know what the next crisis will be.
Scenarios have historically emphasised the previous crisis: severe unemployment, property price declines, credit deterioration. That is a reasonable template and it is not what caused the 2023 regional bank failures.
Those failures involved rising interest rates producing unrealised losses on high quality securities portfolios, combined with concentrated uninsured deposits. Rising rates were not the adverse scenario, since scenarios generally assume rates fall in a recession. The stress that occurred was, in one dimension, the opposite of the one being tested.
The Model Convergence Problem
Because banks are assessed against a common scenario using broadly similar methodologies, their risk models converge.
Convergence means institutions position similarly, and similar positioning means they respond to shocks similarly. A supervisory exercise intended to reduce systemic risk can increase correlation between institutions, which is a genuine and acknowledged tension.
Reverse Stress Testing
The partial answer is to run the exercise backwards. Rather than asking what happens in a defined scenario, ask what scenario would cause failure, and then assess how plausible it is.
This surfaces vulnerabilities that a predetermined scenario misses, because it starts from the institution actual exposures rather than from a supervisor imagination.
It is harder to standardise and therefore harder to use for setting capital requirements across many banks, which is why it supplements rather than replaces the conventional approach.
How to Read Published Results
The headline is the projected minimum capital ratio under stress. More informative is where the losses come from, since that reveals the concentration of each institution exposure.
Comparing the severity of the scenario across years matters too, since results are not comparable if the scenario changed. A bank performing better than last year may be facing a milder test.
The Bottom Line
Stress tests project capital under a supervisor defined adverse scenario and control distributions based on the outcome. They restored confidence when introduced and remain valuable for capital planning. Their limitation is structural: they test the imagined crisis, they encourage model convergence across institutions, and the 2023 failures happened through a channel the scenarios were not built to examine.