Early Warning Model for Financial Risks of Listed Companies Based on Machine Learning

Abstract

The descriptive text information in the annual reports of listed companies is an essential part of the information disclosure of listed companies. The prediction ability of their financial risks can be improved by mining and analysing listed companies’ disclosure text. By extracting the textual characteristics of the “Discussion and Analysis of Business Conditions” in the annual reports of listed companies in the A-share market, we construct textual characteristics indicators that can reflect financially distressed companies and normal companies. Subsequently, the text feature indicators are combined with financial indicator data and classified using convolutional neural networks to construct the financial risk warning fusion model E-CNN. AUC evaluates the performance of the early warning model. The experimental results show that the financial text features extracted by the word2vec-ES model can improve the AUC values predicted by the financial early warning model. The word2vec-ES improves the AUC values predicted by the financial early warning model more significantly compared with other methods, indicating that the word2vec-ES model effectively extracts the financial text features and improves the prediction ability of the financial risk early warning model of listed companies.