The Contribution of Sustainability and Governance Signals to Return on Equity Prediction: Evidence from Tree-Based Machine Learning, Bootstrapped Grouped CV and SHAP

In the global economy, traditional accounting-based ratios alone are often insufficient to fully explain firm performance, increasing the importance of complementary information sources such as sustainability and governance disclosures. In this context, environmental, social, and governance (ESG) indicators, together with corporate governance signals, have increasingly been recognized as important drivers of firm performance. However, the literature does not provide a clear and generalizable view on the impact of ESG indicators on profitability. This study aims to examine whether sustainability and corporate governance signals provide additional information value beyond traditional financial ratios in predicting ROE. To this end, two models were compared using a sample of 428 non-financial publicly traded companies operating in Turkey. The firm-level dataset was constructed using financial statements and independent audit disclosures obtained from the Turkish Public Disclosure Platform (KAP). Tree-based machine learning models were employed to capture potential nonlinear relationships and complex interactions between financial and non-financial indicators. Model performance was evaluated within a Bootstrapped Grouped Cross-Validation framework that considered firm-level dependency; the statistical reliability of performance differences was tested using bootstrap-based confidence intervals and matched tests. Among the evaluated models, Random Forest achieved the strongest overall predictive performance. In conclusion, this study demonstrates that sustainability and corporate governance disclosures provide statistically significant additional information value to ROE prediction. Due to the use of multiple algorithms, it contributes to the literature in a generalizable manner.