Considering the characteristics of volatility with regime switching of the stock market and the misspecification of the parametric model,the paper proposes a nonparametric GARCH model with Markov regime switching,and uses nonparametric estimating technique to estimate the volatility.We classify the volatility of China’s stock market into three states: the fall,the consolidation and the rise.We estimate and forecast the volatility of both Shanghai and Shenzhen stock markets using parametric and nonparametric GARCH models with Markov regime switching,and then we evaluate the performances of these models using MSE1,MSE2 and QLIKE.The result shows that the parametric and nonparametric MRS-GARCH model in which the error term follows a normal distribution is more accurate.Based on these models,we estimate the dynamic VaR of the return of Shanghai composite index and Shenzhen component index of China’s stock market.Then,using Kupiec back testing,we evaluate the performances of these models to predict the market risk.It is concluded that the parametric and nonparametric MRS-GARCH models can both estimate the VaR of China’s stock market well,and what is more,the performance of the nonparametric MRS-GARCH model is better than that of the parametric MRS-GARCH model.