Identifying Green Bond Greenwashing Behavior Using Large Language Models: Evidence from China

Green bonds, as an important financing tool for promoting the transition to a low-carbon economy, warrant attention to their actual environmental benefits. The existence of greenwashing behavior not only harms the interests of investors but may also undermine the credibility of the green financial market. This study takes the Chinese green bond market from 2016 to 2024 as the research object, introduces the BERT artificial intelligence model into the field of greenwashing behavior identification, and constructs a greenwashing identification framework covering two dimensions of selective disclosure and expressive manipulation with six sub-indicators. The study finds: (1) Compared with traditional bag-of-words methods and machine learning methods, the BERT artificial intelligence model has significant advantages in identifying greenwashing behavior; (2) With the improvement of regulatory policies, the greenwashing strategies of bond issuers show dynamic evolution characteristics, the space for selective disclosure is compressed, and expressive manipulation becomes the more dominant form of greenwashing; (3) Bond issuers with greenwashing behavior are more likely to engage in strategic green innovation in the future, and their future financial performance and sustainability are significantly weaker than non-greenwashing entities, and this situation exists in both bond issuers with selective disclosure and expressive manipulation. This study provides a new technical path for the automatic identification of greenwashing behavior in green bonds and provides empirical evidence for regulatory agencies to improve the information disclosure system.