Stock Price Prediction Using Empirical Mode Decomposition Based Theta Method and Forecast Combination

Data-driven decision making and prediction are crucial research area. However, future prediction using past data patterns is very challenging in many cases especially for nonlinear and nonstationary data like stock price data. To improve forecast accuracy, this study focused on decomposition-based hybridization method and forecast combination. Firstly, locally adaptive empirical mode decomposition algorithm and Theta method were used to develop hybridization method. Secondly, affine combination was used to combine forecasts of hybridization method and other method. Developed from hybridization method and forecast combination, proposed method was employed on four stock price datasets of Japanese NIKKEI225 index. Forecast accuracy of the proposed method was compared with six single and hybrid methods using six error measurements. Empirical results revealed the outperformance of the proposed method over the compared methods. Therefore, empirical mode decomposition-based forecasting methods are potentially effective and they deserve further exploration to improve nonlinear nonstationary time series forecasting.