Comparative Analysis of VaR Estimation of Double Long-Memory GARCH Models: Empirical Analysis of China’s Stock Market

Abstract

GARCH models are widely used to model the volatility of financial assets and measure VaR. Based on the characteristics of long-memory and lepkurtosis and fat tail of stock market return series, we compared the ability of double long-memory GARCH models with skewed student-t-distribution to compute VaR, through the empirical analysis of Shanghai Composite Index (SHCI) and Shenzhen Component Index (SZCI). The results show that the ARFIMA-HYGARCH model performance better than others, and at less than or equal to 2.5 percent of the level of VaR, double long-memory GARCH models have stronger ability to evaluate in-sample VaRs in long position than in short position while there is a diametrically opposite conclusion for ability of out-of-sample VaR forecast.