An Empirical Investigation of the VaR of Hedge Funds

Abstract

We compare the performance of several Value-at-Risk (VaR) models when applied to a high frequency hedge fund index. Our analysis is carried out on the Barclay/Calyon CTA daily index available since early 2000. We use 1-day-ahead VaR forecasts for various thresholds (10%, 5% and 1%) and apply univariate and multivariate VaR backtesting procedures. Our results show that the efficiency of VaR forecasts primarily depends on the type of quantiles used for computing VaR forecasts. The choice of the model used to forecast volatility (simple smoothing average, EWMA, symmetric or asymmetric GARCH models) proves much less important in that specific case. Our results also show that the most flexible form is the Cornish-Fisher expansion for 10% and 5% thresholds, whereas Student quantiles are the best to forecast efficiently 1% VaRs.