Abstract: This paper first does non-linear time series test on the SSE Composite Index returns ratio, then carries on the constitutive change examination to the time series, andfinds that the SSE Composite index is the sequence proceeds of non-linear time series as well as structural changes.After Constructing a three conditions, third-order lag's different variance Markov switching model, we do empirical analysis to SSE Composite Index week returns ratio from December 21st, 1990 to August 22nd, 2008, using the maximum likelihood estimation method to estimate the model parameter, and identifying the three main conditions of stock market fluctuations:slo wup, slow down and speed up.Empirical results indicate that the Markov switching model can portray stock market's gradual fluctuation characteristics effectively.

Key words: Markov switching model, China’s stock market, BDS, CU SUM, hetero scedasticity