A novel GA-SVR time series model based on selected indicators method for forecasting stock price

Forecasting stock price is always the hottest topic for investors. In recent years, many time series models have widely been used in forecasting stock price for achieving the smallest lost in investment. However, the previous time series models still have some problems: (1) previous researches selecting the important technical indicators depend on subjective experiences and opinions; (2) conventional statistical models must satisfy assumptions about variables in data analysis; (3) conventional time series models only considered single variable and linear variable; (4) it is difficult to determine the parameters of Support vector Regression (SVR). In order to improve these problems mentioned, this study proposed a novel GA-SVR time series models based on selecting indicators method for forecasting stock price. The proposed model adopted multivariate adaptive regression splines and stepwise regression to select the important indicators. Then, this study constructed the forecasting model by SVR, and used GA to optimize the forecasting model under RMSE. For evaluation the forecasting performance of proposed models, the stock prices of Chunghwa Telecom from 2003 to 2012 years are used as experimental dataset and the root mean square error (RMSE) as evaluation criterion.