Nonlinear Association Between Controlling Shareholders and Financial Reporting Integrity: An Explainable Optuna-Optimized Ensemble Learning Approach in Egypt and Saudi Arabia

Financial reporting integrity (FRI) plays a critical role in capital market efficiency, yet its determinants remain difficult to model due to nonlinear relationships, heterogeneous firm characteristics, and institutional differences across emerging markets. Prior research largely relies on linear econometric approaches, which may overlook threshold effects and complex governance dynamics. This study develops an explainable Optuna-optimized Extremely randomized trees (ET) ensemble framework to examine the association between controlling shareholders and FRI in Egypt and Saudi Arabia. Using a panel dataset of 1746 firm-year observations over the period 2014–2022, the model incorporates advanced preprocessing and mutual information-based feature selection to enhance predictive accuracy and robustness. The proposed model significantly outperforms regularized linear models, standalone machine learning models, and alternative ensemble techniques, achieving R2 values of 0.7935 in Egypt and 0.9231 in Saudi Arabia, alongside substantial reductions in RMSE and MAE. Diebold–Mariano tests confirm that these performance gains are statistically significant (p < 0.01). Explainability analysis using SHAP reveals that firm size and market share are the dominant drivers of FRI, while blockholder ownership exhibits a nonlinear and context-dependent association. Partial dependence results show a complex, non-monotonic relationship in Egypt—consistent with a monitoring–entrenchment trade-off—contrasted with a predominantly positive and monotonic association in Saudi Arabia. Importantly, these nonlinear patterns are not detected in conventional panel fixed effects models, highlighting the limitations of standard econometric specifications in capturing complex ownership dynamics. The findings highlight the importance of institutional context in shaping governance outcomes and demonstrate how explainable ensemble learning can uncover hidden nonlinearities in financial reporting behavior. This study contributes by identifying nonlinear thresholds and cross-country variation in ownership effects while integrating predictive performance with interpretability, offering a robust framework for analyzing corporate governance mechanisms in emerging markets and supporting more informed decision-making by investors, regulators, and policymakers.