A Hybrid Model Combining a Gated Recurrent Unit Network Based on Variational Mode Decomposition with Error Correction for Stock Price Prediction

Abstract

Due to the nonlinearity and high volatility of financial time series, hybrid forecasting systems combining linear and nonlinear models can provide more precise performance than a single model. Therefore, this study proposes a hybrid stock price forecasting model with error correction based on secondary decomposition. The modules of data decomposition, a prediction module and an error correction module constitute the overall framework of the model proposed in this paper. First, variational mode decomposition (VMD) decomposes the original stock closing price data and the initial prediction error sequence. Second, a gated recurrent unit (GRU) neural network model is selected to predict the original subseries. Finally, a new hybrid model consisting of the VMD method combined with an autoregressive integrated moving average (ARIMA) model is established to correct the error subseries. The proposed model is verified through four financial time series. The results show that (a) The combination of the VMD method and GRU network has a better prediction effect than a single model. (b) The combination of the VMD method and ARIMA method can effectively amend the error sequence and improve the prediction accuracy. (c) Error correction to the hybrid model has a great effect on improving the prediction performance of the model. The empirical results have illustrated that the proposed model indeed displays a good performance in forecasting stock market fluctuations compared with other baseline methods.