Non-gaussian Distribution for Var Calculation: an Assessment for the Italian Market

In this paper we compare different approaches to compute VaR for heavy tailed return series. Using data from the Italian market, we show that almost all the return series present statistically significant skewness and kurtosis. We implement (i) the stable models proposed by Rachev et al. (2000), (ii) an alternative to the Gaussian distributions based on a Generalized Error Distribution , (iii) a nonparametric model proposed by Li (1999). All the model are then submitted to backtest on out-of-sample data in order to assess their forecasting power. We observe that when the percentiles are low, all the models tested produce results that are dominant compared to the standard RiskMetrics model. Copyright @ 2001 IFAC.