Improving the accuracy of stock Return Predictions in the Iraqi Stock exchange Index Using Artificial Intelligence Techniques

Account and Financial Management Journal e-ISSN: 2456-3374 Volume 10 Issue 12 December 2025, Page No.-3919-3929 DOI: 10.47191/afmj/v10i12.06, Impact Factor: 8.167 © 2025, AFMJ 3919 Mohammed Abdulrahman Alsendi, AFMJ Volume 10 Issue 12 December 2025 ABSTRACT: This study investigates the application of Artificial Intelligence (AI) and Machine Learning (ML) models to predict stock returns in the emerging Iraq Stock Exchange (ISX), a market characterized by high volatility and limited prior research. The research conducts a comprehensive empirical comparison between traditional statistical models (ARIMA) and advanced AI algorithms, including XGBoost, Linear Regression, Random Forest, SVR, LSTM, and Transformer. Using daily index data from 2018 to 2025, the models were evaluated based on statistical metrics like RMSE, MSE, MAE, and R². The results demonstrate a clear superiority of advanced models over traditional ones. The Transformer model achieved the best performance, exhibiting the highest predictive accuracy and lowest error, attributed to its self-attention mechanism that effectively captures complex temporal dependencies in the data. The findings confirm the research hypothesis that advanced ML models can significantly outperform traditional statistical methods in the challenging context of the Iraqi market. The study concludes by recommending the formal adoption of these AI tools, particularly the Transformer model, to enhance investment decision-making, improve risk management, and foster greater efficiency and investor confidence in the ISX. This research provides a valuable framework for integrating AI-driven financial analysis in emerging markets. KEYWORDS: Artificial Intelligence, Machine Learning, Stock Market Prediction, Iraq Stock Exchange, Financial Forecasting, Empirical Analysis The nature of artificial intelligence (AI) usage in finance and financial markets has changed significantly over the last two decades. They have since then become a staple of forecasting and financial analysis. Classical-styled statistical models based on linear structures and stationary probability density functions (PDF) for returns (Tsantekidis et al., 2022), are now less able to contain, in an equal proportion as data flood into financial market scale upmotion, complex entanglements of nominal features with behavioral profiles. AI and ML algorithms are also better than EMT in curtailing the big data that we can possibly work out non-linear and complexity patterns, in markets helping to enhance stock return prediction accuracy and reducing financial risk (Gu et al., 2020). Stock return prediction is one of the earliest and most challenging problems in finance, since (i) s there are macroeconomic factors including GDP and inflation; behaviorfactors like investor sentiment; (ii) it is a partially efficient market. Classic statistical models (e.g., ARIMA or GARCH) are also used for modelling in current times if only a few data points are available, but they do not capture complicated interdependencies and sudden shifts (Fischer \& Krauss, 2018). Recent studies have shown that AI-based models such as XGBoost (Chen and Guestrin (2016)), long short-term memory network LSTM (Nikolenko et al. Despite that unambiguous evidence, AI is acceleration in leading capital markets (an example is the United States and China), while there's limited research context areas for emerging market such as Iraq society SGX (AlHomaidi et al., 2022). There are also not too many papers in such space using advanced maching and deep learning tehniques but a lot of works with more classical statistical tools. So, there exists an obvious research and empirical gap in the work that no study has yet been submitted using this statistical vs AI breaking point analysis on low efficient & high volatile emerging market context Therefore, this research aims to evaluate the effectiveness of AI algorithms in predicting stock returns in the Iraq Stock Exchange by comparing their performance with classical statistical models like ARIMA. The research also seeks to test the ability of models such as LSTM and Transformer to improve prediction accuracy and to determine their capability to capture dynamic patterns in a highly volatile emerging market “Improving the accuracy of stock Return Predictions in the Iraqi Stock exchange Index Using Artificial Intelligence Techniques” 3920 Mohammed Abdulrahman Alsendi, AFMJ Volume 10 Issue 12 December 2025 “Improving the accuracy of stock Return Predictions in the Iraqi Stock exchange Index Using Artificial Intelligence Techniques” 3921 Mohammed Abdulrahman Alsendi, AFMJ Volume 10 Issue 12 December 2025 “Improving the accuracy of stock Return Predictions in the Iraqi Stock exchange Index Using Artificial Intelligence Techniques” 3922 Mohammed Abdulrahman Alsendi, AFMJ Volume 10 Issue 12 December 2025 “Improving the accuracy of stock Return Predictions in the Iraqi Stock exchange Index Using Artificial Intelligence Techniques” 3923 Mohammed Abdulrahman Alsendi, AFMJ Volume 10 Issue 12 December 2025 “Improving the accuracy of stock Return Predictions in the Iraqi Stock exchange Index Using Artificial Intelligence Techniques” 3924 Mohammed Abdulrahman Alsendi, AFMJ Volume 10 Issue 12 December 2025 Rank Model MSE RMSE MAE R² Interpretation 1 Transformer 0.0001476 0.01215 0.00743 -0.0119 Best model, in terms of accuracy, lowest average error 2 ARIMA 0.0001376 0.03325 0.00922 -0.0119 Good performance, especially in linear time series 3 SVR 0.0007865 0.02805 0.0225 -0.0133 Balanced, good capture of non-linear relationships 4 XGBoost 0.0004919 0.02229 0.01205 -0.0233 Average performance, unstable, risk of overfitting 5 LSTM 0.0005976 0.00299 0.0299 -0.0561 Advanced, but shows relative weakness due to limited data 6 Linear Regression 0.0265716 0.16301 0.029 -181.0913 Simple traditional model, not capturing complex relationships 7 Random Forest 0.0265716 0.16301 0.029 -181.0913 Weakest model, large errors and very low R² “Improving the accuracy of stock Return Predictions in the Iraqi Stock exchange Index Using Artificial Intelligence Techniques” 3925 Mohammed Abdulrahman Alsendi, AFMJ Volume 10 Issue 12 December 2025 Table 3: Comparative Performance of Financial Forecasting Strategies Rank Strategy Final Value Total Return Volatility Sharpe Ratio 1 Transformer 10,739.24 7.39% 0.0000043 170,970.81 2 Linear Regression 75,700.02 6.57% 0.5196 12.64 3 SVR 16,419.67 64.20% 0.0360 17.85 4 ARIMA 10,970.14 9.70% 0.0091 10.63 5 XGBoost 7,963.09 -20.37% 0.3007 -0.68 6 Random Forest 696.61 -93.03% 2.5597 -0.36 7 LSTM 98.60 -99.01% 0.1001 -9.89 “Improving the accuracy of stock Return Predictions in the Iraqi Stock exchange Index Using Artificial Intelligence Techniques” 3926 Mohammed Abdulrahman Alsendi, AFMJ Volume 10 Issue 12 December 2025 Scenario Mean Return Std Dev Max Return Min Return Cumulative Return Sharpe Ratio Positive Days Base Scenario 0.0029% 0.0005% 0.0038% 0.0020% 1.33% ★★★★★ ★ 5.81 100.0% Optimistic Scenario 0.0035% 0.0006% 0.0045% 0.0024% 1.60% ★★★★★ ★ 5.81 100.0% Pessimistic Scenario 0.0023% 0.0004% 0.0030% 0.0016% 1.06% ★★★★★ ★ 5.81 100.0% Stagnant Market Scenario 0.0029% 0.0000% 0.0029% 0.0029% 1.33% 0.00 100.0% High Volatility Scenario -0.0394% 1.4677% 4.6175% -4.7625% -20.49% -0.027 49.67% “Improving the accuracy of stock Return Predictions in the Iraqi Stock exchange Index Using Artificial Intelligence Techniques” 3927 Mohammed Abdulrahman Alsendi, AFMJ Volume 10 Issue 12 December 2025