We apply the machine-learning technique proposed by Zhou et al. (2024) to analyze credit rating reports in China’s bond markets, identifying featured vocabulary and generating text analysis scores. Compared with the traditional bag-of-words text analysis, evidence suggests three advantages of machine-learning scoring. Firstly, it covers featured vocabulary that compensates for missing information; secondly, it reduces misclassification of words’ sentiments; moreover, it mitigates the problem of equal weighting inherent in the bag-of-words method. Our findings indicate that the featured vocabulary neglected in the bag-of-words method plays a crucial role in text analysis and significantly contributes to bond pricing. Additionally, we find that machine-learning text analysis can address AAA rating inflation within China’s bond markets to some extent. In contrast, the bag-of-words method exhibits limited efficacy in mitigating this issue.
Read full abstract