Can ESG disclosure predict carbon risk? Evidence from machine and deep learning models

Abstract

As ESG disclosure becomes increasingly shaped by shifting regulatory agendas, its role in predicting carbon risk warrants closer examination. This study employs machine learning, deep learning, and ensemble techniques to assess whether ESG and financial indicators can effectively predict carbon risk. Results demonstrate that advanced AI models significantly outperform traditional regressions by capturing complex, non-linear relationships often overlooked by conventional methods. SHAP analysis further identifies environmental disclosure as the most influential predictor. These findings offer a timely, data-driven framework for investors, firms, and policymakers navigating the uncertainties of evolving sustainability reporting standards. © 2025 The Authors

Keywords AI predictability, Carbon risk, ESG disclosure, Machine learning