Machine Learning Prediction of Environmental, Social and Governance Reporting Quality: A Global Cross‐Sectional Analysis

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:ML手法の優位性とSHAPによる解釈可能性を確認し、ESG開示研究に新たな方法論的基盤を提供する。

🏢実務担当者:自社のESG報告品質をグローバルベンチマークと比較し、ガバナンス指標の改善点を特定するのに役立つ。

🏛政策担当者:開示品質予測モデルを規制監視ツールとして活用し、グリーンウォッシングリスクの高い企業を特定する手法として注目すべき。

In an era of growing stakeholder pressure and regulatory fragmentation across global jurisdictions, the quality of environmental, social and governance (ESG) reporting has become foundational to corporate valuation, investment screening and regulatory oversight. This study uses machine learning (ML) techniques to predict global reporting quality and examine how the determinants differ in disclosure quality in developed and emerging economies as well as civil and common law jurisdictions. Drawing on a cross‐sectional sample of 5000 publicly listed companies across 50 countries for the fiscal year 2022, we develop and evaluate Random Forest and XGBoost models alongside a panel regression benchmark, using financial performance metrics, corporate governance indicators and institutional characteristics as predictors. The primary contribution of this study is methodological: The authors demonstrate that ML techniques deliver superior predictive power compared to traditional econometric approaches by capturing the non‐linear, high‐dimensional interactions that characterise ESG disclosure decisions globally—a capacity that conventional ordinary least squares and fixed‐effects regressions structurally cannot replicate. The study integrates signalling theory, legitimacy theory and agency theory to explain corporate disclosure motivations across diverse institutional settings. The results show that agency theory, in particular, illuminates why board size and board independence consistently emerge as strong predictors, since independent monitoring reduces information asymmetry and incentivises management to commit to transparent sustainability disclosures. Results confirm the superiority of ML techniques, with XGBoost achieving a test R 2 of 0.78 compared to 0.62 for panel regression. SHapley Additive exPlanations (SHAP) analysis identifies firm size, governance score and board independence as the most consequential predictors. Board independence exhibits a threshold effect: ESG quality gains plateau beyond approximately 65%–70% independent directors. The study offers actionable insights for investors and regulators seeking to identify firms at high risk of greenwashing through governance‐marker profiling and sustainability officers benchmarking their organisations' reporting quality against global peers.

• semanticscholar https://doi.org/10.1002/csr.70732first seen 2026-06-10 05:20:28