Unlocking disclosure narratives: readability, length, and sentiment cues as indicators of bankrupt Japanese companies

Abstract

We investigate potential bankruptcy warning signs in the MD&A and Future Forecast sections of flash reports in Japanese companies. The results show that, compared to solvent companies, companies at risk of bankruptcy tend to provide narrative disclosures with lower readability, fewer sentences, and more negative words. Thematic models (sentiment analyses), regardless of approach (dictionary-based or machine learning), outperform traditional syntactic models (readability and length analyses). This remains true even after considering various control elements such as z-score, company size, company age, industry sector, and years of observation. Our results thus support the inclusion of narrative and thematic (sentiment) analyses in bankruptcy research.