FORECASTING STOCK RETURNS USING RADIAL BASIS FUNCTION NEURAL NETWORKS: EVIDENCE FROM BANK OF BAGHDAD

Abstract

This study investigates the application of Radial Basis Function Neural Networks (RBFNNs) for forecasting stock returns in the Iraqi Stock Exchange, with a specific focus on Bank of Baghdad shares. The research addresses the critical challenge of predicting stock returns in emerging markets characterized by high volatility and limited market efficiency. We employ a comprehensive dataset spanning five years (2019-2023) of daily trading data, incorporating technical indicators, macroeconomic variables, and market sentiment factors as input variables. The RBFNN model architecture is optimized through systematic hyperparameter tuning, including radial basis function selection, network topology, and regularization parameters. Performance evaluation employs multiple metrics including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Directional Accuracy (DA). Results demonstrate that the RBFNN model significantly outperforms traditional econometric models and standard multilayer perceptrons, achieving a directional accuracy of 67.3% and RMSE of 0.0432. The model exhibits superior capability in capturing non-linear relationships and complex patterns inherent in emerging market dynamics. Robustness tests confirm model stability across different market conditions, including periods of high volatility. The findings contribute to the growing literature on artificial intelligence applications in financial forecasting and provide practical insights for portfolio managers and risk assessment professionals operating in emerging markets. The study's implications extend beyond the Iraqi context, offering a methodological framework applicable to similar emerging market environments.