Research On Stock Price Forecasting Based On Support Vector Machine

The future trend of stock prices has always been the core issue of stock market investors. So, do the future price of the stock can be predicted by the use of technical analysis? In order to answer this question, it is to date back to 1970 Fama’s seminal paper, "Capital Markets Efficient", which had established a controversial efficient market hypothesis. The efficient market hypothesis widely questioned yet has become the core object of public research. In the stock market, the tit for tat with the efficient markets hypothesis is technical analysis school, if the stock market is weak form efficient market, any technical analysis, i.e., any attempt to use securities history information to predict future prices of the securities is of no avail of. In the voices of doubt of efficient market hypothesis, the research purpose of this paper is to answer a series of questions: first, whether the stock price or the trend of the future can be predicted with the help of technical analysis. Second, if the future stock price can be predicted, then what methods can be used. Third, what will be the outcome of the forecast, and what kind of conclusions will be obtained.Firstly, the research object of this paper is tomorrow’s closing prices and the opening prices of all constituent stocks of the SSE 50. In this paper, we first review affirmation and negation of China’s stock market’s weak form efficiency market on both sides of the point of view, and it is denied that China’s stock market is weak efficient market in this paper. Then it is supposed that China’s stock market is not weak form efficient market; next, this paper find out eight factors that may affect tomorrow opening price and closing price and that are conducted through correlation tests, finally we find out the Granger reasons of tomorrow’s closing price and the opening price withthe help of the Granger causality test. Next, the epsilon-SVR support vector machine regression method of support vector machine(SVM) of statistical machine learning is taken to predict tomorrow stock opening price and closing price, the kernel function of the SVM is radial basis function(RBF kernel, RBF)(formula 4. 12). Experiments are conducted on the opening and closing prices of the all constituent stocks of the SSE 50.Secondly, as for research method, there are the comparisons between the models in and out of this paper. To be begin with, 5 different models(formula 5.1, 5.2, 5.3, 5.4,5.5) are established and the empirical analysis was carried out on all constituent stocks of the SSE 50 by using these five different models. Through the analysis and interpretation of the empirical results, screening out the prediction model(formula 5.3),that is, using 3 days of historical data to predict tomorrow’s closing price of the average error is the smallest(table 5.17). Next, the model(formula 5.6) was carried out on the second day of the opening price forecasting. Finally, the comparisons between the models of the literatures, respectively, and the reference [20]SVM prediction model and reference BP [55] neural network prediction model to predict.Finally, the following conclusions are drawn: first, this paper found that the use of3 days of historical data to predict the next day’s model 3(formula 5.3) to get better prediction results, the average error is the smallest, 1.936%(table 5.17). This conclusion may be due to the non-steady state and the high noise of financial time series and the momentum effect of stocks’ price, so that the price trend of the stock can be extended.Second, through the Grainger causality test the paper found that the history of trading volume is not the Grainger reason of the next day’s closing price(table 4.2). Third,empirical results of the next day’s opening price found that model 1(formula 5. 6) on 48 stocks of the SSE 50(except for the two stock delisting China North Vehicle and the Oriental Pearl) and the prediction mean hit rate of the ups and downs reached 75.58%(table 5.23), far higher than the random walk hit rate of ups and downs of 50%,indicating that China’s stock market has not reached the weak form efficient market.Fourth, this paper found that the forecast for tomorrow’s close shows a certain lag, and in the future day opening price of the forecast, there is no similar phenomenon. Fifth,the prediction model of this paper is better than the reference SVM [20] prediction model(section 5.5), and it is better than the reference literature BP [50] neural networkprediction model(section 5.6).At this point, the three questions that have been raised at the beginning of the abstract have been answered in the affirmative. Therefore, this paper has the following significance. First of all, in theory, this paper will answer the question of the effectiveness of China’s stock market, and provide research reference for the researchers.Secondly, in practice, this paper is tempt to provide some reference information for investors.