The stock market can be defined as a market that, on the one hand, facilitates companies that need financing and, on the other hand, provides opportunities for investors who need to invest. By predicting the rise and fall of stock indices, it can bring guidance to individuals and companies when to enter the financial market, and it can also provide theoretical implications for government economic policy making. However, the stock market is a complex system full of various information, it is not only affected by past information, but also by current political, economic and psychological factors, so it is difficult to accurately predict the rise and fall of the stock index. At present, the stock index rise and fall prediction methods are mainly applied technical analysis method and measurement time series analysis method, which applied technical method is used by more groups, because it almost does not need too much analysis but according to personal investment habits and experience, subjective color. The econometric time series method is a method that is effective only when used in an ideal situation, which requires the input of the independent variable indicators and the target variable is preferably linear, if it is a non-linear situation, the results will have no reference significance. In this paper, we combine the main capital flow model with support vector machine as a tool to construct a stock index up/down prediction scheme.