A hybrid model for stock price based on wavelet transform and support vector machines

This study presents a hybrid data mining model for stock price that combines wavelet transform and support vector machines. In our proposed method, the wavelet transform is firstly applied to eliminate the noise of the stock time series. Secondly, the stepwise regression is employed in feature selection. Thirdly, the time delay concept is utilized to obtain the optimum prediction model. For illustration and evaluation purposes, this study refers to the results of empirical testing conducted on several stock markets, including the HS300 index and the stocks of five financial institutions in the China´s security market. The features of predictor include 34 financial statement variables and six news variables in stocks. Empirical results demonstrate that the proposed model consistently and significantly outperforms the single support vector machine as well as the conventional logistic regression and neural network. Consequently, we prove that the wavelet transform technique effectively eliminates noise from stock time series, and that the stepwise regression technique effectively reduces dimensionality.