6533b7d4fe1ef96bd1261c3a

RESEARCH PRODUCT

Non-Gaussian Distribution for Var Calculation

subject

description

Publisher Summary This chapter compares different approaches to computing Value-at-Risk (VaR) for heavy tailed return series. Each model has been submitted to a backtest analysis. The most representative asset returns of the Italian stock market and the exchange rates for the major currencies are used. The results obtained confirm that when the percentiles are below 5%, the hypothesis of normality of the conditional return distribution determines intervals of confidence whose forecast ability is low. In fact, it is observed that the return distributions are asymmetric and leptokurtic and the hypothesis of normality is usually rejected when subject to statistical test. Among the alternative models proposed, the α-stable densities and the generalized error distributions are reliable in the VaR calculation. The generalized error distribution approach is not suitable for large portfolios and the numerical procedure to compute the stable density percentiles is quite complex.