Capability Comparison of the Models based on Long Memory and Dynamic Neural Network Models in Forecasting the Stock Return Index in Tehran Stock Exchange

The aim of this study is to introduce an efficient nonlinear model for predicting the return of Tehran Stock Exchange (TSE) Price index. For this purpose, the daily time series of price index from Farvardin 1388 to Aban 1390 is used. This study includes 616 observations; 90% of which used for estimating coefficients and the remaining 60 observation are deduced for out of sample forecasting. By comparing the results of a nonlinear dynamic artificial neural network (NNAR) and a nonlinear regression model (autoregressive fractional integration moving average «ARFIMA»), we found that NNAR models have better performance in out of sample forecasting based on mean square error criteria (MSE) and root mean square error criteria (RMSE) than the nonlinear regression models (ARFIMA).