Abstract: In this paper, skewed generalized error distribution (SGED) is introduced to account for skewed and heavy-tailed financial asset returns, and SV-SGED model is proposed to model asset return volatility, and then dynamic value-at-risk (VaR) can be measured. In order to test the accuracy of risk models, the back-testing technique is adopted. At the same time, a method for maximum likelihood (ML) estimation of SV models is introduced based on the efficient importance sampling (EIS) technique. Finally, an empirical study of Shanghai Stock Exchange composite index is presented. Empirical results demonstrate that the SV-SGED model can describe asset return volatility better than the SV model based on normal distribution (SV-N) and the SV model based on generalized error distribution (SV-GED), and the SV-SGED model can yield more accurate VaR estimates than the SV-N and SV-GED models.

Key words: VaR, SV models, skewed generalized error distribution, efficient importance sampling, maximum likelihood estimation