Abstract: This paper constructs a model of ETF fund return based on the double exponential jump diffusion process and two-factor jump diffusion process, and presents the solving method for parameters based on MCMC. According to the four kinds of asymmetric GARCH model, it establishes a time-varying EVT-POT-GPD method to further measure quantile of standard residuals of ETF fund return, and presents a method of determining the dynamic market risk VaR and CVaR. The empirical results prove that using two-factor model and one-year historical data can better forecast the price of ETF fund, the return of ETF fund market suffering different impact from good and bad news has significant asymmetry characteristics and leverage effects, and the TGARCH model is better in measuring positive asymmetric shocks and conditional heteroscedasticity. It finds using the EVT-POT-GPD method constructed to calculate standard residuals of ETF fund return, with the improvement of confidence level, threshold value k will move towards tail and quantile will increase gradually. Additionally, dynamic CVaR of the five ETF fund is greater than VaR obtained by any GARCH model, and when the confidence level rises from 95% to 99%, the CVaR growth rate is far faster than the growth rate of VaR. It also shows that the sensitivity of CVaR in measuring risks is much higher than that of VaR. Meanwhile, back-testing and the chi-square test find that the success rate of CVaR and VaR rises with increasing the confidence level.

Key words: fund, risk, diffusion, asymmetric structure, dynamic