A New Approach to Comparing VaR Estimation Methods


Journal of Derivatives

hiver 2008, vol. 16, n°2, pp.54-66

Départements : Finance, GREGHEC (CNRS)

Mots clés : Value-at-Risk, Bank Trading Revenue, Backtesting, Coverage Test

Value-at-risk (VaR), despite its known shortcomings, has become established as the most commonly used measure of risk exposure. But many variants of procedures for implementing VaR exist. Some variants use historical data with or without simulations, while others assume parametric models, such as GARCH, with parameters estimated from past data. And, of course, different users might focus on different VaR cutoffs: 5%, 1%, and so on. Perignon and Smith use an innovative method of extracting daily values for bank revenues from their annual reports to explore which VaR methods empirically work best. A second innovation discussed in the article is how to measure the accuracy of tail estimation at multiple points in the tail. The results suggest that, in estimating VaR for banks, parametric methods work best