The U.S. has no shortage of data on the causes and effects of the Great Recession. Regulators have drawn on that information in designing rules intended to curtail the possibility of a repeat event. The severely adverse scenario used by the Federal Reserve's stress tests, for example, mimics the economic data observed during that dark period.

But no two recessions are identical. Though a complete autopsy of the 2008-9 financial crisis is desirable, the real challenge for stress-test designers will be to envisage the next recession and ensure that banks are sufficiently capitalized to withstand the vagaries that such an event will throw at bank capital buffers.

I have no special insight as to what this event might look like. But other major recessions in my living memory suggest that banks must be prepared to face a wide array of difficulties.

The four most recent downturns — those of the early 1980s, early 1990s and early 2000s as well as the Great Recession — were all distinct in nature, cause, and magnitude. As a result, stark differences emerge in the performance of different kinds of loans across the three most recent recessions. (The early 1980s recession, unfortunately, occurred in a relative data vacuum — but anecdotal evidence suggests that credit performance was not as bad as might have been expected, given the brutal downturn in the real economy.)

The 1990s recession saw prodigious loss rates in commercial real estate and high loss rates for commercial and industrial loans, while household credit performed better. Move forward to the early 2000s recession — triggered by a stock market collapse and usually written off as a minor blip — and the story changes considerably. Commercial real estate, the prime villain in the 1990s downturn, posted no appreciable movement in credit losses. Retail credit performance was solid, with only a moderate rise in credit card losses.

The one category with outsize losses, given the modest severity of the event, was commercial and industrial loans. Imagine a recession with the same narrative as 2001, but of the same magnitude as 2008-9, and the potential losses for commercial and industrial loans would be staggering.

Of the three most recent recessions, the 2008-9 crisis is the only one in which credit losses were appreciably elevated in all loan categories.

An analysis of the data suggests that if the proximate cause of a recession is related to an activity financed by banks, the sector in question will incur massive losses. Excesses in commercial construction were a primary cause of the 1990s recession, and therefore losses were highest in the commercial real estate sector. A housing market collapse caused the Great Recession, and credit losses were highest in residential real estate. Meanwhile, sectors that are quiet on the eve of recessions do not necessarily suffer any stress.

From the perspective of stress-test modeling, several challenges emerge from this analysis. It is straightforward to develop a model that captures the high loss rates observed in commercial real estate in 2009 and 2010. But one wonders whether the same model, employed using data through 2000, would predict the nonexistent commercial real estate losses of 2001. It might also fail to predict the outsize rise in credit losses seen in the early 1990s recession.

Analysis of the 1990s downturn poses perhaps the biggest challenge for stress tests. The overall economy during that period was not that bad, yet commercial loan losses induced widespread bank failure. This suggests that a bank might be sufficiently capitalized to withstand a severely adverse economic scenario — but still fail in a less severe, more narrowly targeted, hypothetical event. Such scenarios, those involving a single market segment for instance, usually unfold as a result of excessive lending or poor industry underwriting standards during specific asset market booms.

Current stress-testing protocols were clearly designed with the failures of Washington Mutual and Lehman Brothers in mind. Ideally, the process should also be able to deal successfully with a potential repeat of the 1980s savings and loan crisis and various other kinds of events.

The application of the severely adverse scenario should not lull bank executives and regulators into thinking that banks will necessarily withstand lesser recessions, such as that of 1991 or even 2001. History suggests that there is no such thing as a perfectly safe bank.

Tony Hughes is managing director of credit analytics at Moody's Analytics, where he manages the company's credit analysis consulting projects for global lending institutions.