A Stock Price Forecasting Method Using Autoregressive Integrated Moving Average model and Gated Recurrent Unit Network

Stock price predict is a challenging problem because stock data has the characteristics of high noise, dynamic, nonlinear and non-parametric. In this study, we proposed a new stock price forecasting method using autoregressive integrated moving average (ARIMA) model and gated recurrent unit (GRU) network. Firstly, the stock sequence is processed by the differential operation to stabilize the stock sequence. Secondly, the residual of each stock sequence is obtained by using the ARIMA model. Thirdly, according to the prespecified threshold of residual, each stock sequence is divided into training sets which are of different sizes. Finally, the GRU network is used to predict stocks price. The experimental results show that the accuracy of our method for predicting stocks price is better than that of recurrent neural networks, long-short term memory and gated recurrent unit.