Direct versus iterated multi-period Value at Risk

Although the Basel Accords require financial institutions to report daily predictions ofValue at Risk (VaR) computed using ten-day returns, a vast part of the literature deals withVaR predictions based on one-day returns. From the practitioner point of view, some ofthe conclusions about the best methods to estimate one-period VaR could not be directlygeneralized to multi-period VaR. Consequently, in the context of two-step VaR predictors,we use simulated and real data to compare direct and iterated predictions of multi-periodVaR based on ten-day returns assuming that the conditional variances of one-period returnsfollow a GARCH-type model. We show that multiperiod VaR predictions based on iteratingan asymmetric GJR model with normal or bootstrapped errors are often preferred whencompared with direct methods that are often biased and inefficient.