Red-Blooded Risk: The Secret History of Wall Street

Stress Tests

The first tool is stress tests. Like VaR, these are widely misunderstood outside the risk management profession. People think of them as worst-case scenarios. That would be pointless. To paraphrase John Maynard Keynes, in the worst case, we are all dead. You can postulate some plausible extreme worst case if you want, and compute the effect on your institution, but what good is the answer?

A proper middle-office risk management stress test does not postulate a situation and compute the loss; it starts from a level of loss and asks what kinds of events could cause it. The loss level is typically set at three to 10 times the VaR amount. This is the region beyond everyday experience, but not so far beyond that no one can predict what might happen. You then ask what kinds of plausible events might cause such a loss: how big a stock market crash, how severe a liquidity squeeze, how disruptive a natural disaster, how big a bankruptcy of a financial institution or sovereign default? Stress tests are intended to examine the potential for sudden events to cause damage, so you don't ask about more slowly developing crises like recessions.

The people who think stress tests are worst-case scenarios also think the institution should keep enough cash around to pay for projected stress losses. One problem with that is cash does not solve all problems. Cash in the vault is no protection against the vault being robbed. Another problem is that the level of the stress event is arbitrary. ...