The Fed has just announced a second round of stress tests for US banks. This follows the first round of tests in Spring 2009 with results announced in May. There are many interesting issues one could touch on in regard to the role of stress tests in regulating the banking system. But this blog is focused on risk management at non-financial corporations, and so we will leave most of those issues for discussion in other venues. But risk management at non-financial corporations obviously employs many of the same tools as risk management at banks and other financial institutions, so there are a few interesting questions that are common to the two realms. One of them is “What is special about stress tests?”
Are they any different than the other quantitative tools employed to measure risk? If the Fed is already calculating a VaR for each bank, it would seem that it already has a more comprehensive piece of information than would be contained in a stress test. The stress test identifies a single bad scenario and calculates the bank’s losses in that scenario. When calculating its VaR, a bank is effectively determining its losses under many, many possible scenarios. The bank then generates a probability distribution of losses. The VaR just locates the tail on this distribution. What extra information could the Fed extract by forcing the banks to focus on a single scenario instead of the full probability distribution of scenarios?
VaR has many advantages: it is simple to compute, mathematically elegant and relies on readily available market based information. Thus, VaR caught on in the financial industry, and rapidly became the state-of-the-art method of risk management. But, over the years, it became apparent that VaR has many faults. The best known problem is that the distribution of outcomes does not resemble the well behaved Normal distribution. In particular, the likelihood of tail events is much larger that what the Normal distribution would indicate. Six sigma events appear to be a lot more frequent than previously thought. Equally important, downside and upside risks are not symmetrical. The frequency of downside risk seems higher -the distribution exhibits negative skewness-, and the consequences of downside events seem to be much more serious. Most fund managers seem to understand this second aspect, for they focus primarily on maximum drawdowns instead of on standard deviations.
In times of crisis, many of the assumptions lurking behind a VaR calculation are no longer valid. Liquidity dries up for many securities, and the prices calculated from standard models don’t match the terms on which the securities can actually be sold. A crisis can change all of the market dynamics. Managers who control the VaR calculation can manipulate the assumptions loaded into the program to generate a comfortable outcome, and it may be difficult for supervisors to pierce the veil.
All this means that VaRs are indeed very weak measures of true risk. They might give a decent reading under normal conditions, but they are of little use in extraordinary circumstances. Hence, VaRs need to be complemented with other measures of exposure. This seems to be what stress tests try to accomplish: to characterize times of extreme duress, and imagine what might happen if during those times something really bad happens. Focusing on a single scenario is a valuable management tool. As such, stress tests are no different from good planning. As General Eisenhower, an expert who gained his stars fighting in WWII, once put it: “a plan is nothing; planning is everything”.
Managers in many fields have used scenario analysis, the basis of stress tests, for a long time. However, due to their lack of statistical rigor, evidenced by the fact that the scenarios are not weighted by their estimated likelihood, scenario analysis has largely been ignored by the finance profession. It’s time to correct that practice, both in the management of banks, and in the financial management of non-financial corporations.