Do ESG Scores and Controversies Improve Corporate Financial Distress Prediction? Evidence from Machine Learning in Asia-Pacific Markets
Purpose: This study investigates whether ESG information provides incremental predictive value for corporate financial distress beyond established accounting-based predictors. Focusing on twelve Asia-Pacific markets, we evaluate the complete LSEG/Refinitiv ESG framework, including the headline ESG score, pillar and category scores, and the ESG Controversies score, within a rigorous machine-learning framework. Design/Methodology/Approach: The analysis is based on 47,253 firm-year observations spanning 2016–2024. The benchmark model incorporates the principal predictors established in the financial distress literature, including traditional accounting ratios, liquidity measures, funds-flow indicators, earnings-manipulation variables, and macroeconomic conditions. Model performance is evaluated using a strict out-of-time validation design in which models are trained on 2016–2022 observations and tested exclusively on 2023–2024 data. Gradient boosting and complementary machine-learning algorithms are employed, while SHAP analysis is used to examine feature contributions. Findings: The results reveal three principal findings. First, ESG coverage is highly selective, encompassing only approximately one-fifth of firm-year observations and declining to negligible levels in frontier markets, while distressed firms are substantially less likely to receive ESG ratings than financially healthy firms. Second, although ESG variables possess modest standalone predictive ability, they provide no incremental improvement once comprehensive financial fundamentals are incorporated into the prediction model. Third, SHAP attribution assigns considerable importance to ESG variables despite their negligible contribution to out-of-time predictive performance, highlighting that feature importance should not be interpreted as evidence of incremental predictive value. Furthermore, the U-shaped relationship between ESG performance and financial distress documented for U.S. firms is not observed across Asia-Pacific markets. Practical Implications: This study provides the first comprehensive out-of-time machine-learning assessment of the full LSEG/Refinitiv ESG architecture for corporate financial distress prediction across twelve Asia-Pacific economies. Originality/Value: By evaluating ESG information against one of the most comprehensive benchmarks of classical distress predictors, the study demonstrates that ESG ratings currently offer limited incremental value for financial distress prediction in the region, while identifying insufficient ESG coverage as a fundamental constraint for both practitioners and policymakers.