Environmental, social and governance (ESG) rating prediction using machine learning approaches

The study's objective is to predict the environmental, social, and governance (ESG) ratings of firms. Applying six machine learning algorithms, we collect a global data sample of 6166 firms in 73 countries from 2005 to 2019. We use firm-specific and macroeconomic predictors in the model and find that Random Forest Classifier provides the highest accuracy (78.50%) among the six machine learning algorithms by considering Kappa, area under the curve, receiver operating characteristic, and logLoss. The variable importance factor reveals that the lagged ESG score has the highest contribution to the model. Firm size has the second highest, and debt to equity ratio has the third-highest contribution, which indicates that a firm's total assets and a firm's financial leverage impact the ESG rating. In addition to a contribution to the growing body of ESG literature, the study's findings can help practitioners, firm regulators, and policymakers in social and environment-friendly decision-making and investors in investment decisions.