Forecasting credit ratings of decarbonized firms: Comparative assessment of machine learning models

Abstract

• We assess the ability of machine learning models to predict credit ratings. • For low-carbon firms, classification and regression trees have the best precision. • The random forest ensembles have the second-best accuracy of rating predictions. • In firms with financial distress, artificial neural network models are also suitable. Maintaining low carbon energy transitions is a phenomenon that is critical in curtailing greenhouse emissions. However, such shifts usually warrant incremental capital expenditures, which require an uninterrupted access to financing. Credit ratings are an essential consideration of the financing process. In this paper, we assess the ability of various machine learning models, in order to forecast the credit ratings of eco-friendly firms. For this purpose, we have employed a sample of 355 Eurozone firms that are ranked on the basis of the extent of their climate change score by SDP, between the years spanning from 2010 to 2019. The study uses various machine learning methods, and the findings suggest that classification and regression trees have the most precision for the credit rating predictions. Even when the forecasting was constrained to the investment grades, speculative grades, or default categories, the accuracy remained robust. The results also suggest that a random forest ensemble can be used alongside the regression trees in order to predict default or near default ratings. Given that such firms face dynamic risk exposure towards environmental, ecological, and social factors, these results have important implications that can be taken into consideration when assessing the credit risk of pro-ecological firms.