Examining Environmental, Social and Governance Role in Predicting Early Default Warning: A Machine Learning Perspective

Abstract

The study has examined how Environmental, Social, and Governance (ESG) performance can predict early warning in defaulting in an emerging market through the application of the advanced machine learning methods. The data used in the analysis is firm-level panel data (including 11,000 defaulted and non- defaulted firms across the emerging economies in the time frame 2010-2023 that is obtained through the use of the Refinitiv database. Combining ESG composite scores and pillar-specific indicators with traditional financial distress metrics, the study elaborates a holistic early warning framework in the determination of whether sustainability performance is informing on conventional accounting-based, as well as market-based risk measures. The study has been methodologically executed through various supervised learning algorithms such as the random forest, Extreme gradient Boosting (XGBoost), Support Vector machines models. The empirical evidence suggested that that the use of ESG measures enhances predictive accuracy compared to financial ratio-based models to a large extent. The quality of governance turns out to be the strongest predictor of early default risk, then exposure to environmental risk, and social factors are also heterogeneous across the industries. Machine learning models achieve better performances than the traditional econometric specifications in measuring nonlinearities and the complex cross-dimensional interactions. The research adds to the literature in the field of financial distress because it introduces sustainability risk in credit risk modeling in emerging markets. It has methodological insights in the form of interpretable machine learning methods and practical implications to investors, regulators, and credit risk practitioners who are interested in incorporating ESG data into early warning and risk management systems.