Portfolio crash testing: making sense of extreme event exposures

ABSTRACT The topic of extreme events is becoming ever more important for risk management. Stress testing is a technique that is explicitly designed to deal with extreme shocks; however, its methodology and place in the risk process is often unclear to risk managers. This paper addresses some common misconceptions about stress testing and provides methodology for its incorporation into the risk process as a supplement to risk measures such as value-at-risk and tracking error. Two stress testing models are presented and empirically validated on actual extreme periods. Both are based on multivariate normal distributions conditional on a factor shock, differing only in the way that covariances are estimated. The first model uses temporal weighting commonly used in the risk model construction; the other uses event weighting, which assigns a higher weight to extreme events that are similar to the factor shock specified. The key conclusion is that the time weighted model performs better in moderate or semi-expected shocks, while the event weighted model performs better in more extreme and unexpected shocks like the Long-Term Capital Management crisis, 9/11 terrorist attacks, and the Fall 2008 financials-led meltdown. The event weighted model, which is designed to reflect the rise in correlations and variances during extreme markets, produces a more conservative estimate of return impacts. Our results support the conclusion that stress testing can be a very valuable addition to standard risk measures.