Stock Index Prediction Based On ARIMA-GARCH And SVR

With the advent of the big data era,the understanding of financial markets has been deepened.How to predict financial data has become an active research area involving economics,statistics and computer science.The profound description of financial market fluctuation starts from the ARIMAGARCH model.However,as one of the traditional time series analysis models,ARIMA-GARCH has certain deficiencies in explaining the nonlinear factors.And these factors can be detected by machine learning methods,such as support vector machines,as they require fewer assumptions on data.In other words,these machine learning methods can resolve the linear limitations of traditional methods.Therefore,machine learning is commonly used to process the nonlinearity and the fluctuation of financial markets.Taking account of the advantages of ARIMA-GARCH model on processing conditional heteroscedasticity of time series data and the power of support vector regression(SVR)when processing nonlinear data,this paper proposes an SVRARMA-GARCH model,which combines ARIMA-GARCH model and SVR model.It is a combination of traditional time series analysis and modern machine learning methods.Meanwhile,this paper proposes another combined model,called GARCHSVR model,to conduct comparison experiments between several models.This paper uses ARIMA-GARCH,SVR,GARCH-SVR,and SVR-ARMAGARCH model to predict the stock index and test the performance of the models.In addition,as feature selection is an important part of SVR,this paper applies stock technical indicators to analyze index and feature selection algorithm to collect the variables in SVR.Then multiple sets of stock data are used to train and evaluate the four models.The experimental results show that:(1)The SVR-ARMA-GARCH model has the highest accuracy among the four models,while the accuracy of SVRARMA-GARCH model with SVR prediction results has significantly improved compared the ARIMA-GARCH model in terms of MAE and RMSE.(2)Feature selection in SVR,especially the GARCH-SVR model with GARCH term,can improve the performance of SVR models.