Chinese Stock Prices Research Based On Support Vector Machine

The stock market itself is a very complex system with nonlinear and chaotic properties. Itdoes not comply with the efficient market theory (EMH) which is universal applied in thefinancial markets. Besides, the stock price is effected by too many factors, such as nationalpolicy, the macroeconomic situation. So the stock market is very difficult to predict, and it is thedifficulty and hotspot in financial fields. China’s stock market is a very special stock market. Ithas just a short history, but it is still not completely synchronized with the mature stock marketsof foreign countries, and it even exists some phenomenons which financial economics can notexplain. Stock is too important for China’s national economy and people’s life, so its effectiveforecasting has a very important practical significance.The study finds that, although the stock market can not be predicted in a long period, it isfeasible for the short-term trend prediction. There are some methods to predict and analysis thestock market. Compared with the traditional forecasting methods, artificial intelligence methodsobviously have more advantages and prospects. As one of the artificial intelligence methods,support vector machine (SVM) method can solve nonlinear and small sample problems, so it ismore suitable for stock market predictions than other methods.The research goal of this paper is to establish a SVM prediction model of Chinese stockprices. The paper took the stock market prediction as the research object, then chose six relevantindicators which have important impacts on the stock price forecasting and used SVM to predictthem. For the chaotic characteristics of stock market, the paper firstly used SVD noise reductionmethod to preprocess the data, then used the phase space reconstruction technology. Through theC-C algorithm to find the optimal embedding dimension and delay time, the original series isembedded in a high-dimensional space, so that more comprehensive properties of the system areshown out and its orderliness is increased. Use the GP algorithm to verify whether the system ischaotic, through which to ensure the correction of using phase space reconstruction step. And forthe two key technical problems of SVM---parameter optimization and kernel function selection,the paper used PSO optimization algorithm to optimize parameters, and then combined two kernel functions with better predictive kernels according to a certain weight. By adjusting theweight, the paper built a new mixed kernel function which is the best and most suitable for stocksystem.For the validity of the prediction model, the paper chose six indicators from420tradingdays of Shanghai index to do the simulation experiments. Realize the model step by step and testthe experiment results to ensure that every step is optimal. The prediction accuracy of the modelis showed through the experiment results. Experiment results show that the Shanghai index timeseries is indeed chaotic, and phase space reconstruction, PSO optimization algorithm andmixture kernels all have improved the prediction accuracy. The stock price prediction modelbuilt by these techniques has very excellent forecasting effect and it is feasible and effective.