Structural Breaks Of Share Indices Returns And Modeling Volatility In China Stock Markets

Chinese stock market, an emerging market with only 20 years' history, is apt to be influenced by various outside factors and shows a great fluctuation in result of our imperfect market mechanism, lagged legal system and immature psychology of investors and so on. Therefore, the intensive study of its fluctuations becomes especially important, and many scholars delve into modeling volatility for share indices returns.The development of modeling volatility has gone through three major stages: the early traditional econometric model under the assumption of fixed variance, later AutoRegressive Conditional Heteroskedasticity (ARCH) model and Stochastic Volatility (SV) model, and newly developmental non-parameter model for high-frequency data. Now the most widely used model is still (G)ARCH-type models, but in the practical application, this model could be combined with structural breaks in time series.On this basis, this paper applies dummy variable of structural breaks in GARCH-type models, aims to fit the return series of Shanghai Composite Index and Shenzhen Component Index (which from Dec. 16, 1996 to May 31, 2010). The content mainly contains two aspects: Firstly, detect the structural breaks in variance from two sample series with ICSS algorithm, the essence of which is to construct suitable statistics with a series of recursive residuals, and then simulates its distribution and critical value for hypothesis test. Secondly, put the change-points as dummy variables into GARCH model to re-fit them, pose a contrast of goodness of fit, degree of forecast in all circumstances to select the optimal one. There are extra factors need to be considered, including risk premium, asymmetric effect and error distribution, etc.The empirical studies of sample indicate that parts of significant events in Chinese stock market have caused variance of share indices returns some structural breaks. Considering these breaks, it is better to make them as dummy variable to join EGARCH(1,1) model, in which, goodness of fit for t-distribution error is higher while degree of forecast for GED is better. And then, further study would measure VaR of asset income more accurately.