Bankruptcy Prediction from 10-K Narratives: Evidence from Interpretable Text Scores and Accounting Baselines

The authors investigate whether narrative disclosures in annual 10-K filings provide "incremental" warning signals for corporate bankruptcy that are not captured by traditional quantitative accounting ratios. Specifically, they seek to determine if a custom-built, distress-specific text score can improve the accuracy of bankruptcy prediction models beyond standard accounting baselines and existing general-purpose financial sentiment dictionaries.

The study constructs a Pre-Bankruptcy Stress (PB Stress) Score based on five thematic pillars: liquidity/funding stress, debt/covenant/refinancing stress, operating deterioration, restructuring/legal distress, and business fragility. The researchers extracted text from Items 1, 1A, and 7 of 10-K filings—sections where management typically discusses business risks and financial conditions.

To evaluate the score's effectiveness, the authors compared it against:

• A Five-Variable Accounting Baseline: A model using standard metrics like firm size, leverage, liquidity, profitability, and operating cash flow.
• The Loughran-McDonald (LM) Dictionary: A standard benchmark for financial sentiment that measures broad negative or uncertain tone.

The analysis used a sample of 40,475 firm-year observations (2010–2021) from the SEC EDGAR database, matched with bankruptcy events from the Florida-UCLA-LoPucki Bankruptcy Research Database.

The PB Stress Score significantly enhances predictive performance. When added to the accounting baseline, the model’s Area Under the Curve (AUC) increased from 0.8323 to 0.9019. Furthermore, the model’s ability to capture bankruptcies in the top-decile of risk rose from 44.12% to 64.71%. These results remained robust across various validation tests, including out-of-time samples and bootstrap inference. The authors conclude that distress-specific narrative language provides unique, interpretable information that acts as an early warning system before financial deterioration is fully reflected in quantitative accounting statements.