Forecasting Framework Using Hybrid Modeling and Support Vector Regression

Abstract

Time series forecasting for financial market has increasingly attracted the interests of investors and academic researchers. In recent years, some hybrid models have been constructed to improve the predictions since some methods cannot extract useful information from stock time series with noise to conduct prediction. In this research, a prediction framework is proposed to forecast the stock market behavior using the methods of wavelet coherence, multiscale decomposition and support vector regression (SVR). First, a combined method is applied to the raw data and remove noise to get useful information. Then, a SVR model is applied to improve the prediction performance of multidimensional nonlinear data. Furthermore, the comparison experiments were performed with both Shanghai Composite Index and Dow Jones Index to examine the effectiveness of the framework. The results indicate that the proposed framework performs better than other advanced models.