Signal or noise? The role of firm narratives in earnings prediction

This study examines whether narrative disclosures in corporate financial reports enhance the prediction of future firm performance. While quantitative financial data such as current earnings and cash flows are established predictors, the added value of narratives remains uncertain. Using machine learning and topic modeling on over 5,000 10-K filings, we test whether narrative features improve predictive accuracy. Results indicate that narratives add limited and inconsistent value, though sections like Management Discussion and Risk Factors have higher predictive power. Our findings underline both the promise and current limitations of integrating textual analysis into financial prediction models.