Could another financial crisis on the scale of 2008 happen again? And if it did, how well would today’s banks withstand the shock? To answer those questions, government regulators regularly stress-test large financial institutions, assessing how their balance sheets would perform under extreme economic conditions. A recent Federal Reserve Board stress test found that even in a severe recession, the largest U.S. banks could collectively absorb losses of $708 billion while remaining financially sound and able to continue lending.
Most stress tests rely on historical crises or hypothetical scenarios that closely resemble them. However, new research from the McCombs School of Business at The University of Texas at Austin suggests that the most memorable market crashes are not always the best indicators of future financial risk. Instead, the researchers propose a broader, data-driven method for identifying the scenarios most likely to expose weaknesses in financial institutions.
The new approach, developed by Rui Gao, associate professor, and Stathis Tompaidis, professor, in the Department of Information, Risk, and Operations Management, uses multifaceted market data rather than focusing primarily on headline-making events. In experimental tests, their models outperformed an existing regulatory approach by more effectively identifying the scenarios that produced the largest losses. “It’s not necessarily the large market moves, the headline days, that are the best days to use,” Tompaidis says. “It’s days where the stresses are complemented with each other.”
Stress testing has a long history. Tompaidis notes that the concept dates back to the 17th century, when gunsmiths “proof tested” gun barrels by firing them with heavy loads to see whether they would fail. Modern financial regulators use a similar principle, exposing institutions to hypothetical economic shocks to evaluate how their profits, losses, and capital positions would respond under pressure.
Designing those scenarios, however, is challenging. Regulators must balance transparency with effectiveness. Revealing too much about how scenarios are selected could allow financial institutions to tailor their portfolios to perform well on specific tests rather than genuinely improving resilience. Because stress testing is also costly and time-consuming, each scenario should be carefully chosen to reveal the greatest potential vulnerabilities.
Existing stress tests often centre on well-known market crashes, when stock prices, interest rates, currencies, commodities, and market volatility all shifted dramatically. But Gao explains that real-world risks are more complex. Different financial stressors can move in different directions, meaning several famous crises may actually test institutions in similar ways. A better approach is to combine severe yet complementary scenarios that uncover different types of risk.
Working with former McCombs doctoral student Rohit Arora, the researchers evaluated 2,828 historical market scenarios spanning April 2008 to June 2019. Their algorithm selected four scenarios with complementary stress factors and tested them against 1,000 simulated investment portfolios. The researchers’ scenarios identified the single worst historical outcome about 40% of the time, while their most accurate model successfully captured the five worst outcomes approximately 95% of the time. Compared with the Commodity Futures Trading Commission’s baseline scenarios, the new approach consistently detected more severe losses.
The findings suggest that combining multiple, complementary market stressors provides regulators with a more accurate picture of financial risk than relying on famous market crashes alone. The approach could also reduce the number of stress tests needed while improving their effectiveness. Ultimately, the researchers hope their models will help regulators better identify financial vulnerabilities before the next crisis arrives. As Gao puts it, “We want to choose scenarios to help regulators to evaluate risks more accurately.”
More information: Rohit Arora et al, Choosing Scenarios to Estimate Resilience and Stress Test Financial Institutions, Management Science. DOI: 10.1287/mnsc.2024.06126
Journal information: Management Science Provided by University of Texas at Austin