Combining seasonal time series ARIMA method and neural networks with genetic algorithms for stock price index forecasting

摘要

An accurate prediction method for the stock price index can provide useful information to the investors in order to yield them high returns than others. Most stock price indexes are data with trends and seasonality. Many methods such neural networks and time series methods such seasonal autoregressive integrated-moving average (SARIMA) model can deal with data on trends and seasonality. Artificial neural networks are capable for prediction, but it is difficult to decide what input data are and design good network structure. Based above, this paper proposes a hybrid forecasting model. This model combines the SARIMA model and neural networks with genetic algorithms. This paper inputs the analytic result generated by the SARIMA model as the input data of neural network and use genetic algorithms to design optimal neural network structure to develop a new method. Genetic algorithms are used to globally optimize the number of neurons in the hidden layer and learning parameters of the neural network architecture. This study constructs a predictive model by combining SARIMA model and neural networks with genetic algorithms. This model was employed to forecast seasonal time series data of TSEC Taiwan 50 Index in Taiwan stock market. Several procedures were utilized to evaluate forecasts, MAE, RMSE and Wilcoxon signed-rank test. Results in this study can provide a valuable reference for researchers.