How well stress tests measure a bank’s ability to survive adverse conditions depends on the statistical modeling approach used. Banks can access data on loan characteristics to precisely estimate individual default risk. However, macroeconomic scenarios used for stress tests—as well as the reports banks must provide—are for a bank’s entire portfolio. So, is it better to aggregate the data before or after applying the model? Research suggests a middle-of-the-road approach that applies models to data aggregated at an intermediate level can produce accurate and stable results.