The Role of Sentiment in Credit Rating: Natural Language Processing Machine Learning Approach
摘要
We analyse 65,936 English news articles sourced from Factiva and apply the FinBERT model to evaluate how market-wide sentiment towards green bonds affects corporate bond credit rating dynamics. Using fixed-effects regressions that control for both bond-and firm-level characteristics, we find that negative sentiment has a significant adverse effect on corporate bond credit rating outcomes. The results remain robust and consistent across multiple model specifications and comprehensive sets of fundamental controls, highlighting the role of market sentiment in shaping credit ratings within the green bond market. Furthermore, we find that green and conventional bonds exhibit similar risk profiles under comparable corporate financial conditions. In addition, the issuer's ownership type, whether publicly listed or privately held, appears to be associated with credit risk. These findings underscore the importance of soft information, particularly public sentiment, in explaining variations in corporate bond credit risk and contribute to a broader understanding of how market perception influences credit rating assessment in green finance.