Estimation of value at risk by extreme value methods (Q5942928)

The article is devoted to the estimation (prediction) of quantiles of financial assets returns distributions. Such quantiles are called Values at Risk (VaR). The author describes methods based on the Gaussian model, empirical quantiles, estimation of parameters of generalized extreme value and generalized Pareto distributions (GPD). To include in the model possible dependence and heteroscedasticity of the returns the author propose to use a GARCH model with heavy-tailed disturbances. The parameters of the GARCH model are estimated using a Gaussian-model likelihood and then the distribution of the residuals is fitted by GPD. The quantile of the fitted distribution is taken as the VaR. These methods are applied by the author to a portfolio consisting of the bank-stocks ``Den Danske Bank'' and ``Jydske Bank'' by the data from 1985 to 1998. The author conclusion is that the GARCH+GPD technology is the one that describes the tail features of financial time series the best.