Predicting financial distress using textual risk disclosures in annual reports: How and what risks are disclosed?
A study of 48,224 U.S. annual reports from 2006 to 2023 finds that analyzing the language of risk disclosures can predict financial distress better than traditional models using only numbers. The way risks are described and the specific risks mentioned both matter, especially for long-term forecasts. As time horizons lengthen, text analysis outperforms numerical data. The research urges investors and regulators to focus on risk language for early warning signs, though results may differ outside the U.S.
What it examines
This study examines if the way and type of risks described in annual reports can help predict financial distress in U.S. firms. Using text analysis and topic modeling on over 48,000 reports from 2006--2023, it compares the value of textual risk disclosures to traditional financial and market data.
What it concludes
The results show that risk disclosures in text improve predictions, especially for longer-term forecasts, more than traditional data. This suggests investors and regulators should pay close attention to these disclosures. Future research could further explore how different types of risk topics affect financial distress predictions.
Evidence objects
Analyzing risk disclosure text in 48,224 U.S. annual reports (2006--2023) significantly improves financial distress predictions, outperforming traditional models that rely solely on numerical financial, market, and macroeconomic data.
key_findings bullet 1 · key_findings · validation V0
The study finds that both the way risks are described and the specific types disclosed offer valuable forecasting insights, with textual analysis becoming more useful than numbers as prediction horizons extend further into the future.
key_findings bullet 2 · key_findings · validation V0
A novel approach combining textual attributes and topic modeling highlights the importance of disclosure language, urging investors and regulators to use these early warnings, though findings may differ for non-U.S. firms or regulations.
key_findings bullet 3 · key_findings · validation V0
This paper innovatively applies textual attribute analysis and topic modeling to 48,224 annual reports, revealing that textual risk disclosures surpass traditional numerical variables in predicting financial distress, especially over longer horizons. Its dual focus and scale offer fresh insights, advancing quantitative risk management, financial AI, and NLP applications in finance.
key_findings bullet 4 · key_findings · validation V0
Raw abstract and provenance
Textual risk disclosures in annual reports, which directly and foresightedly discuss firms’ potential risks negatively impacting their operations, are rarely considered in financial distress prediction. This study explores whether textual risk disclosures can provide valuable information to improve financial distress prediction accuracy. To comprehensively extract information from textual risk disclosures, textual attributes are utilized to capture “How” risks are disclosed, and a topic model is adopted to identify the textual topics to reveal “What” risks are disclosed. Based on textual risk disclosures from 48,224 annual reports of U.S. firms from 2006 to 2023, the empirical results demonstrate that incorporating risk disclosures improves predictive performance compared to the benchmark using numerical financial, market, corporate governance, and macroeconomic variables. Moreover, “What” risks are disclosed can offer more information than “How” risks are disclosed. Last but most importantly, as prediction horizon increases to a longer time, numerical variables show a significant decline in predictive ability, whereas textual risk disclosures can provide more helpful information and even improve prediction performance better. This study enlightens that investors and regulators should pay attention to textual risk disclosures in annual reports when assessing financial distress risks.
Source row: 1569 · abstract type: unknown