A Study On Short-term Prediction Of Shanghai Composite Index Based On SVM

Stock market is a complex non-linear system, and is affected by many factors. In order to ensure the interests of stock investors and reduce the investment risk, we need to predict stock market fluctuations and then grasp the law of development of the stock market. The domestic and foreign researchers have established many outstanding models to forecast the securities price, for example: Time sequence model, ANN model, GA model and so on. These forecast methods regard and solve securities price forecast question from different perspective, and their forecasting result have certain value. However these methods have its respective limitation, cannot achieve good results in the reality application. Relative to other models, SVM model solves problems which exist in other machine learning methods, such as over-learning, due to learning, local minimum, curse of dimensionality and so on. SVM model has now been successfully applied in the domain of securities prediction.First, the relevant background knowledge of the stock market is introduced. Existing methods of forecasting the stock market are introduced, and we point out the inadequacies of forecasting methods. Then we comprehensively introduce the statistics theory of learning and support vector machines method, and describe the basic principle of the support vector machines method in detail.Next, the support vector machines method is used in the stock market forecasting, and we proposed the basic flow during the stock market forecast. During the test we use Shanghai Composite Index as a research object and basic indicators as input variables. According to closing price and trading volume from different period, we design 9 projects. Through 9 groups of comparative trials, we index the test result and discover which project has the highest precision. Finally we determine the time validity of the closing price and volume in stock price forecast.Third, the support vector machine model is designed. We select 4 kinds of technical indicators as input variables, and design 2 projects according to the data from different stages of pre-designed process. The input variable of project 10 is series of technical indicates which left in relevant analysis. Through making the factorial analysis to the input variables of project 10, we decrease the dimension of the input variable effectively, and extract the principal components. The project 11 takes the principal components which extracted by factorial analysis as input variable.Finally, compare with the 11 groups of prediction results. We draw the conclusion: 1. During the SVM method forecasting process, the fundamental indicator compared to the technical indicator be able to enhance forecast precision, when selects the input variable; 2. The closing price and the trading volume as input variables have 3 days of time validity, during the SVM prediction process.