Multi-step-ahead Stock Price Index Forecasting Based On Hybrid Models

Forecasting stock market price index has been regarded as one of the most challenging application of modern time series forecasting since the stock market is a complex, evolutionary, and non-linear dynamic system.Recently support vector machine (SVM) based on the Structure Risk Minimization (SRM) principle which seeks to minimize an upper bound of the generalization error rather than minimize the training error have successfully solved prediction problems in many domains. However, the behavior of stock price index cannot easily be captured by individual model, such as SVM model. Therefore, a hybrid framework is a good alternative for forecasting stock price index.This paper hybridizes SVM with the self-organizing maps (SOM) technique and particle swarm optimization (PSO) algorithm to solve multi-step-ahead forecasting of stock price index. The hybrid framework conducts the following processes: SOM algorithm to cluster the data automatically according to their similarity to resolve the problem of insufficient training data; SVM models for regression are built on the categories clustered by SOM separately; and PSO to optimize hyper-parameters of SVM models. The multiple SVM models provide various patterns in predicting the stock market, so the hybrid framework is able to cope with the fluctuation of stock market values. The proposed framework was demonstrated using a real dataset--Dow Jones Industrial Average to predict the closing price index of next five trading days, moreover, two variants of multi-step-ahead prediction strategies were compared. The experiment results show that the proposed framework is an improvement over the traditional single SVM based on the criteria of normalized mean square error (NMSE) and mean absolute percentage error (MAPE), and Diebold and Mariano test.