Prediction Model of Stock Market Returns Based on Wavelet Neural Network

Abstract

The traditional prediction model is not able to achieve a satisfying prediction effect in the problem of a non-linear system and nonstationary financial signal. The existing wavelet neural network has overcome the deficiency of traditional prediction model which is limited to linear system when predicting. However, wavelet neural network has a defect of confusing signal frequency. Based on the theory of wavelet analysis, the paper designs wavelet neural network that has eliminated confusing of signal frequency by using improved single sub-band reconstruction algorithm. In this paper, weight adjustment and learning of network adopt improved weight adjustment algorithm and Levenberg-Marquardt algorithm respectively. It takes returns in Shanghai stock market from January 10th, 2006 to July 18th, 2008 as example to compare simulation error of stock market returns between BP network and wavelet neural network. The results show that the simulation result of improved wavelet neural network is more accurate than that of BP network, and wavelet neural network constructed in the paper can forecast stock market returns.