Integrating Text Analytics and Financial Indicators for Corporate Risk Early Warning

Abstract

Public financial reports contain large amounts of qualitative information that is difficult to incorporate into traditional risk assessment models. Manual analysis is costly and inconsistent across analysts. This study examined 36,800 annual and quarterly reports published by listed companies between 2015 and 2022. Textual indicators related to liquidity pressure, litigation exposure, and management uncertainty were extracted using hybrid statistical and semantic methods. These indicators were combined with financial ratios to construct early warning models. Results show that adding text-based features improved default prediction accuracy by 8.4% on average. In high-volatility sectors such as real estate and manufacturing, improvements exceeded 11%. Misclassification remained common for firms undergoing restructuring, indicating limitations in automated interpretation. Text analytics provides complementary risk signals rather than a replacement for traditional financial evaluation.