Prediction of Stock Market Returns with out of Sample Data: Evaluating out of Sample Methods (Regression Method and Wavelet Neural Network)

Understanding stock price behavior and price forecasting are one of the most challenging issues that speculators، traders and brokers are faced with. Since stock market was established in the nineteenth century، many researchers have focused on investigating the stock price forecasting model. At the outset some of the models such as Autoregressive، ARMA، ARIMA، ARCH and GARCH were widely applied. Although some of these models were better than others in performance، but none of them had satisfactory result. Recently a number of researchers have been considering the stock market as a dynamic nonlinear system. On the other hand، artificial intelligence techniques including neural networks، genetic algorithms and fuzzy logic have produced successful results in solving complex problems. In this study we tried to use wavelet transform and neural network to present a model that more accurately predict the return on stock market index. In this hybrid model، the smoothing property of wavelet transform is used to reduce the noise level of data then a neural network model is used for forecasting. Comparison of forecast error in ARIMA models، neural network and wavelet neural network indicates that noise reduction improves forecasting of the index returns.