Investigating corporate default risk from carbon risk and ESG perspectives using explainable artificial intelligence

Abstract

In response to the severe challenges posed by global warming and extreme weather events, China has established the “dual carbon” goals through a comprehensive policy framework that integrates economic development with environmental sustainability. Against this backdrop, Chinese listed companies, as key economic players, have become the main force in carbon emission reduction. This study explores key factors influencing financial default risk of listed companies in China based on the Explainable Artificial Intelligence (XAI) framework. The study selects indicators from six dimensions: corporate carbon emissions, environmental, social and governance (ESG) scores, distance to default, management discussion and analysis (MD&A), investor sentiment, and financial indicators. To address the category imbalance problem due to the scarcity of samples of special treatment (ST) companies, three techniques—random Oversampling, adaptive synthetic sampling (ADASYN), and synthetic minority oversampling (SMOTE)—are comprehensively tested, and the performances of six machine learning models are evaluated, with results showing that the prediction effect of Explainable Boosting Machine (EBM) and eXtreme Gradient Boosting (XGBoost) is optimal. By introducing Shapley values for global, local, and interaction visualization analysis, not only was the order of feature importance identified as follows: financial indicators > MD&A > ESG > positive investor sentiment > corporate carbon emissions, but also the cut-off points of FAC1 (−0.8385), MD&A (0.0405), and ESG (6.21) within the financial indicators were discovered, providing a quantitative basis for assessing corporate sustainability.