Research On Stock Index Forecasting Based On Linear And Nonlinear Support Vector Machines Combined Model

As an important part of financial market,stock market has been paid close attention by investors and scholars all the time.Therefore,it is of great theoretical and practical significance to analyze and predict the trend of the stock market.Many scholars have put forward many prediction models or methods,such as time series analysis,neural network model and so on.However,stock market is a dynamic complex nonlinear system.With the development of the market and the innovation of information technology,the limitations of earlier models are becoming more and more obvious.Therefore,there is an urg ent need for new analytical methods and techniques to be proposed.Because the stock market in our country has both linear characteristics and nonlinear characteristics.Therefore,constructing a model that contains both linear and nonlinear features is more suitable to fit the stock market.First of all,this paper selects opening price,closing price,the highest price,lowest price and turnover of Shanghai composite index as input variables,the next day's closing price as output variables,respectively using the linear support vector regression model and RBF kernel support vector regression model to do regression prediction of Shanghai Composite index.Then,in order to overcome the shortage of incomplete information extraction of single model,this pap er adopts the optimal weighting method to combine the two single models above.Finally,this paper proposes a combination for correcting error,that is using linear support vector regression model to forecast the Shanghai index,then extracted the unexplained residual part as output variables,still use the most expensive,the highest price,lowest price,opening price,closing price and turnover of the Shanghai composite index as input variable,using nonlinear radial basis kernel support vector regression model to extract the information of the residual part again.And then,combine the results of the two models as the final output.This study finds that the prediction effect of the optimal weighted combination model is not very ideal.Although the predict ion error is smaller than the linear support vector regression model,but is bigger than the radial basis kernel nonlinear support vector regression model,this paper argues that the reason for this result is that the two single models have similar theoret ical basis and mathematical models.It cannot achieve the ideal goal that using mutual advantages to make up for each other's shortcomings.The combination model of error correcting in each index reflects the superiority,reflecting that single models cannot fully extract the forecast information of stock market.To correct single models by extracting prediction errors of single models can achieve better prediction effect.