A comparison of GARCH with ANN model for forecasting stock index returns
by
Mahmoud Helan
Department of Marketing &Management, University of Bahrain

This paper considers whether artificial neural networks (ANN) model can outperform GARCH time series models for forecasting stock index returns.. This paper involves the ANN models using Back-Propagation algorithm. Further, using output from AR-GARCH model as input to a neural network is explored. Several procedures were utilized to evaluate forecasts, RMSE, MAE and encompassing test. The Jordanian stock index returns are studied using data starts from January 3, 1998 through to December 29, 2004 a total of 1705 observations. Hundred observations were reserved for forecasting performance evaluation using one step-ahead prediction. The results suggest that artificial neural networks are superior to GARCH models and volatility derived from the AR-GARCH model is useful as an input to a neural network.

Date received: October 24, 2005