Predicting Corporate Bankruptcy: Integrating Financial Ratios and Risk Disclosure Text

Abstract

This thesis investigates whether firms’ narrative risk disclosures can improve the accuracy of bankruptcy prediction models traditionally based on financial ratios. While financial indicators have long been used to assess distress risk, they may overlook forward-looking signals embedded in managerial language. To address this, the study incorporates textual variables extracted from the risk factor section (Item 1A) of 10-K filings into a logistic regression framework. The analysis is based on a sample of U.S. and Canadian public firms from 2014 to 2019, using Compustat financial data and SEC filings. Textual features, including tone, specificity, and references to legal risk are constructed using dictionary-based methods. The results show that these narrative indicators contribute modest but consistent improvements in predictive performance, particularly in out-of-sample tests. The findings suggest that Item 1A risk disclosures contain useful information and can complement traditional models. Future research may benefit from applying more advanced natural language models to uncover deeper signals within corporate text.