The paper is concerned with tail estimation for financial return series and，in particular， the estimation of market risk such as value at risk VaR or the expected shortfal1．W e fit GARCH—models to return data using pseudo maximum likelihood and use a GPD approximation suggested by extreme value theory to model the tail of the distribution of the innovations．W e find that a conditional approach that models the conditional distribution of asset returns against the current volatility background is better suited for VaR estimation than an unconditional approach that tries to estimate the marginal distribution of the process generating the returns．In the empirical research we choose Shanghai stock-market index and find that ES is more efficient than VaR．

Download