Federal stress tests improve, offering more confidence in bank resilience

Could the next systemic shock look like 2008—or something new? Regulators increasingly rely on stress tests to probe how big banks might absorb losses while continuing to lend. A recent Federal Reserve exercise estimated that, in a severe recession, large institutions could lose $708 billion without threatening solvency or credit supply. Yet this approach still leaves an open question: which “what-if” scenarios actually expose the most dangerous vulnerabilities?

Traditional stress testing often centers on iconic market days, such as historic crashes, when prices, rates, exchange values, commodities, and volatility can surge together. The problem is redundancy. If multiple scenarios push the financial system in nearly the same way, they may fail to reveal worst-case outcomes that emerge from less famous—but strategically complementary—combinations of risk factors.

New research from the McCombs School of Business at The University of Texas at Austin proposes a broader, data-driven way to pick stress scenarios. Led by Rui Gao and Stathis Tompaidis (with doctoral graduate Rohit Arora), the method uses multifaceted information rather than betting everything on headline events. The aim is to select scenarios that are both consistent and transparent for oversight, while still limiting room for institutions to “optimize to the test.”

In experimental evaluations, the researchers’ scenario sets outperformed an existing regulatory baseline in identifying extreme losses. As Tompaidis notes, the most informative days are not necessarily those with the biggest headline moves, but those where stresses combine in ways that interact with portfolios’ risk exposures.

Their framework treats scenario selection as an optimization problem under constraints typical of real-world testing. By emphasizing worst-case likelihood, it seeks a small set of high-impact scenarios that collectively stress different dimensions of profit-and-loss dynamics.

Instead of four famous days moving in lockstep, the researchers assemble scenarios so that key risk factors move in different directions—reflecting the complex relationship between market variables and portfolio outcomes in practice. “They cause stress in different ways,” Tompaidis explains, making complementary shocks more revealing than repeated narratives.

They tested the approach against 2,828 historical market scenarios spanning April 2008 to June 2019, selecting four scenarios with complementary stress profiles. Across 1,000 simulated portfolios, the new set captured the single worst historical outcome about 40% of the time. A more accurate version identified the five worst outcomes about 95% of the time.

Compared with baseline scenarios used by the Commodity Futures Trading Commission (CFTC), the proposed selections produced more severe loss estimates. The takeaway is straightforward: combining stresses intelligently can model risk more effectively than replaying familiar Black Friday-style episodes.

The research argues that such scenario design could reduce the number of stress tests needed while improving the probability of detecting tail outcomes. Ultimately, the goal is to help regulators assess resilience more accurately before the next crisis.