CSI 300 Index Regression Prediction Based On Support Vector Machine

Stock maraket has dual characteristics of both high-risk and high-yield, its price fluctuations have always been an important indicator of our country’s economic development level, and it’s also the focus of investors and the gov-ernment, therefore the prediction of stock price is particularly necessary. But the stock price fluctuations are highly nonlinear, the accuracy is very low if we establish mathematical prediction model through traditional methods.In the 1990’s, as a kind of machine learning theory based on statistic learn-ing theory and structural risk minimization principle, support vector machine (SVM) has been putting forward. In solving the problems of machine learn-ing, high dimension and nonlinear, SVM shows excellent properties; In terms of nonlinear, by using kernel function instead of inner product calculation in the higher dimensional space, SVM masterly converts nonlinear problems into linear problems in the higher dimensional space.Considering the advantages of SVM in solving nonlinear problems, we apply it in the problem of CSI300 index stock price regression, and do research on stock price fluctuations based on SVM from two different aspects.On the one hand, we establish SVM stock price regression prediction mod-el based on optimization algorithm, the model mainly use the day before’s opening price, highest price, closing price, lowest price, trading volume and turnover to predict the day’s opening price. The selection of kernel function’s parameters in SVM has a great influence on the accuracy of model’s prediction, so we use grid search method, genetic algorithm and particle swarm optimiza-tion to optimize the parameters of SVM’s kernel function, through comparative analysis, we obtain CSI 300 index regression prediction model based on genetic algorithm and SVM.On the other hand, we establish SVM stock price prediction model based on fuzzy information granulation, and in this model, we use time as the in-dependent variable, the daily opening price as the dependent variable, and then do fuzzy information granulation based on time series, every information granule contains three quantities:lowest, average and maximum amount, and then establish SVM regression prediction model based on lowest, average and maximum to predict the range and change of trend of CSI 300 index.With the application of SVM in predicting CSI 300 index, we have achieved desired effect, and in this paper, we not only establish CSI 300 prediction mod-el based on genetic algorithm optimization and SVM, enriching the content of CSI 300 index prediction research, but also establish CSI 300 index predic-tion model based on fuzzy information granulation and SVM, make a further research from trading band’s aspect, which can be more clear to forecast the stock price.