The Comparison Study For VaR Estimating Model In Chinese Stock Market If Long Memory Exists

VaR (Value at Risk) is a statistical technique used to measure the market risk. The survey done by some authoritative financial institutions show that since the 1980s last century, VaR received a great concern and has been employed by many commercial banks, investment banks, non-financial companies, institutional investors and regulators. Many financial institutions seem VaR as a guard against financial risks in the first line of defense, and they developed some software with VaR technique to manage risk. Regulators use VaR technology as a tool for financial regulating, for example in the core principles for effective supervision of banking in Basel issued by the Basel Committee and the EU capital adequacy bill, VaR is an important tool for the supervision of market risk.As the GARCH family model can be used to describe the dynamic characteristics of stock profits and capture cluster effect, nonsymmetrical features in the stock market. Therefore, in recent years many parameter method for calculating VaR focus on combining various GARCH model and the t- distribution, GED distribution which could capture the thick tail feature of the stock market profit.In this paper we first introduce the concept and calculation methods of VaR, and the accuracy test of VaR model; then we recalled the development of traditional GARCH volatility model to FIGARCH model, the concept and test methods of long memory is presented later. In the chapter for empirical analysis, a long memory test is applied first to detect the existence of long memory in volatility for Shanghai Stock Index, Shenzhen Component index and Hongkong Hangseng Index. Strong evidence of long memory in volatility is found. Then the GARCH (1, 1), FIGARCH (1, d, 1) and FIEGARCH (1, d, 1) models are applied to three market indices to assess each model in estimating VaR at various confidence levels. The empirical results show that Fractionally integrated GARCH with GED error model performs the best in estimating 5 percent VaR and FIEGARCH with GED error model performs the best in estimating 1 percent and 2.5 percent VaR.We found FIGARCH and FIEGARCH model performs the best in estimating VaR of China's stock market than the GARCH model. This shows that to consider non-symmetry and long memory for the estimation of VaR in China's stock market is very valuable. Considering long memory features help to enhance the predictive ability for volatility model and the distribution selection has a significant impact on model's predictive capability.