Long-term Investment Analysis Of The Chinese Stock Market Based On SVM

The thesis is about the investment analysis of the Chinese stock market based on Support Vector Machine (SVM). The research focuses on finding out stocks of long-term investment value in Chinese stock market. All listed companies from 1999 to 2011, except ones of Telecommunication Services Industry and Financial Industry, are chosen as study samples. Models are built by sub-industries. For the data is unbalanced, the thesis tries three methods to build a proper model. They are SVM, WSVM (Weighted Support Vector Machine) and Over Sample. The experiment results show that WSVM has obvious advantages in dealing with unbalanced data in this case. As we know, the long-term value of stocks is often associated with fundamental information highly. So the research chooses 23 financial indicators of listed companies as input vectors of the model. Also, because the investment value of stocks is ultimately reflected in the rate of return on their investment, the research chooses the ranking of stocks" 5-year rate of return as label variables. Top 25% of rate of return classified as "+1" class; others are "-1" class. Thus, a stock selection problem turns to a classification problem. In the course of study, five study periods are divided, respectively:1999-2007, 2000～2008,2001～2009,2002-2010,2003-2011. Each study period is used as training data to train models and the period and subsequent periods are used as test data to test models. The final experimental results show that the WSVM model works best in selection of good stocks.