Research On Stock Price Forecasting Methods Based On Support Vector Machines

Stock price forecasting is a hot issue of financial forecasting. With the development of artificial intelligence and computer technology, the stock price forecasting methods have developed largely. Especially, various theoretic achievements based on machine learning and support vector machines appeared endlessly, some of which had been applied in practice and performed well.In this thesis, after introducing statistical learning theory and support vector machines algorithm, some problems such as variable selection, kernel function selection and rule extraction of current research are analyzed. To solve these problems, three stock price forecasting methods based on support vector machines are proposed. In order to illustrate the validity of the proposed methods, these methods are applied to predict stock prices coming from Shenzhen stock market. The main contributions of the thesis are as follows:1 Each research contributions and developments of stock price forecasting methods are summarized. Some problems and difficulties in the field of stock price forecasting based on support vector machines are pointed out. An overview on support vector machines is given. Support vector classification algorithm and support vector regression algorithm are concentrated.2 Three stock price forecasting methods based on support vecor machines are proposed. Aiming at the difficulty to variable selection, a two-stage support vector regression approach is proposed. According to the problem of kernel function selection, two combining forecsting methods are presented. In order to overcome the blackbox problem of support vector machines algorithm, a method which combines support vector machines and rough set is introduced.3 Numerical examples based on Shenzhen stock market are given to illustrate the validity of the proposed methods. Experimental results indicated these methods achieved satisfying results.4 Finally, a conclusion and future research directions are presented.