A new wavelet-neural network-ARIMA shares index combination forecast model

Abstract

To improve the accuracy of forecasting stock prices, a new stock index prediction approach is proposed, which based on wavelet analysis combines the autoregressive integrated moving average (ARIMA) and artificial neural network. The non-stationary share price index series are decomposed and reconstructed into one low frequency signal and several high frequency signals by wavelet; the approximate stationary low frequency signal is predicted using ARIMA forecasting model, and the high frequency signals are forecasted using Elman neural network models; the prediction result of each layer are mixed by the radial basis function(RBF)neural network and the result is the final prediction. Examples show that the prediction of the combined forecasting model is precise.