Could the 2008 financial crisis strike again? How would today's banks react? One way for government regulators to find out is to stress-test large financial institutions: analyze how their balance sheets would weather possible calamities.
A recent Federal Reserve Board stress test found that in a severe recession, large banks could lose $708 billion without imperiling their finances or ability to lend.
Stress tests often draw on historical crises or fictional ones that mimic them. But what if the most famous or intuitively alarming scenarios are not the ones that reveal the deepest financial risks? New research from the McCombs School of Business at The University of Texas at Austin offers a more broad-based approach.
The new models, devised by Rui Gao , associate professor, and Stathis Tompaidis , professor — both in the Department of Information, Risk, and Operations Management — rely less on big events and more on multifaceted data.
In experimental runs, their test scenarios outperformed a regulator's existing ones at identifying worst-case 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."
Performance Under Pressure
The concept of stress testing goes as far back as gunsmiths in the 17th century who would "proof test" their wares, Tompaidis says. "They would fire barrels with heavy loads to see whether the gun would fail or not."
Today's financial stress testing is similar. A regulator provides a financial institution with hypothetical high-pressure scenarios. The institution assesses their effects on its profits and losses.
But designing such what-ifs can be tricky, Tompaidis says. On the one hand, regulators often don't explain why scenarios get chosen. On the other, too much transparency is like telling students in advance what's on their test. Institutions can manipulate the process by choosing portfolios that can better withstand a particular stressor.
Because tests are costly and time-consuming to run, they should also focus on worst cases: scenarios likely to generate the biggest losses.
To design an algorithm for choosing such scenarios, Tompaidis and Gao — with McCombs doctoral graduate Rohit Arora — threaded several needles. They aimed for tests that would capture worst cases, be consistent and transparent, but not allow financial institutions too much leeway.
Multiplying Stresses
Existing stress tests tend to use "famous days," such as historical stock market crashes, or similar hypotheticals created by regulators and economists. On such a single day, stock prices, interest rates, currencies, commodities, and market volatility may all move together.
But instead of moving together, stress factors can move in different directions, Gao points out. "In the real world, the relationship between the profit and loss and the risk factors could be very complicated."
Thus, four famous days might be redundant, if all four of them stress institutions in essentially the same way. A more informative set would include scenarios that are severe but also complementary. Each one exposes different possible weaknesses.
"They're not only causing a lot of stress in the market, but they also cause stress in different ways," Tompaidis says.
The team ran its models against stress tests designed by the Commodity Futures Trading Commission (CFTC). It considered 2,828 historical market scenarios, from April 2008 to June 2019, and selected four that included complementary stress factors.
Across 1,000 simulated portfolios, the researchers' set of scenarios identified the single worst historical outcome approximately 40% of the time. Their most accurate model picked out the five worst outcomes about 95% of the time.
Compared with the CFTC's own baseline scenarios, the researchers' scenarios captured more severe losses.
The lesson is that combining stressors can model financial risks more effectively than reliving Black Fridays, Tompaidis says.
The approach could also reduce the number of tests, he adds, because only a few complex scenarios should be enough to capture the worst cases.
The researchers hope regulators will take note. After all, their goal is to lessen the impact of future crises by better assessing today's risks. Says Gao, "We want to choose scenarios to help regulators to evaluate risks more accurately."
" Choosing Scenarios to Estimate Resilience and Stress Test Financial Institutions " is published in Management Science.
