Early warning of corporate financial crisis based on sentiment analysis and AutoML

Abstract

Establishing an early warning model for corporate financial crises is important for managing risks and ensuring the continued stability of the capital market. A financial crisis early warning indicator system for listed companies was constructed, which includes financial indicators, management indicators and annual report text tone features. Using techniques such as web crawlers and text sentiment analysis, we collected data related to 820 listed companies in mainland China from 2017 to 2021. Six models were then constructed and their results were compared. The results of the comparative analysis showed that: there is room for AutoML to be applied and explored in this area; the model performance and inference speed of integrated learning CatBoost are substantially improved compared with traditional methods; feature importance rankings help to understand the formation of corporate financial distress. Thus, textual information such as corporate annual reports can help predict financial crises.