Testing the machine learning model LSTM in enhancing the accuracy of predicting returns of common stocks: An analytical study of companies listed on the Iraq Stock Exchange for the period (2016-2024)

Abstract

Predicting stock returns represents one of the fundamental challenges in financial markets, particularly in emerging environments characterized by high volatility and low efficiency, such as the Iraqi Stock Exchange. This study aims to evaluate the ability of companies of the Long Short-Term Memory (LSTM) neural network to predict the monthly stock returns of companies listed on the Iraqi market, using monthly data for a sample of 22 from different sectors for the period 2016–2024. The study employed a quantitative analytical approach that included data preprocessing, model construction, and LSTM training according to deep learning requirements, while assessing predictive accuracy using MAE, RMSE, MAPE, and MSE metrics.

he results indicate that the LSTM model achieved solid predictive performance, with a Mean Squared Error (MSE) of 0.0003588, reflecting a small gap between actual and predicted values. The Root Mean Squared Error (RMSE) reached 0.0189420, suggesting that the average deviation between predictions and actual returns does not exceed approximately 1.9%, demonstrating the model’s ability to capture temporal patterns in a market characterized by instability.

The study recommends adopting artificial intelligence techniques, particularly deep learning models such as LSTM, as supportive tools for portfolio analysis and financial decision-making. It further emphasizes expanding the application of such models in the Iraqi financial markets to enhance forecasting quality and improve risk management.