Multi-Perspective Fusion Graph Model for Financial Distress Prediction of Listed Companies

This study introduces a multi-perspective fusion graph model that integrates textual and tabular data to predict financial distress in listed companies. The model constructs separate graphs for textual and tabular data using the KNN algorithm and employs the GraphSAGE model to extract and fuse features from these graphs. The study focuses on Chinese listed companies from 2001 to 2020, utilizing annual report texts, financial data, macroeconomic data, and non-financial data. The results demonstrate that incorporating textual information significantly enhances the predictive capability of the model, with the GraphSAGE model outperforming traditional models such as Logistic Regression, Random Forest, and Artificial Neural Network. The study also highlights the importance of independently extracting features from different data types to improve prediction accuracy. The economic benefits of different models are analyzed, showing that the GraphSAGE model can fully identify low-risk enterprises and achieve higher potential gains under short position strategies. The study concludes by suggesting future research directions, including incorporating additional textual data sources, developing dynamic network structures, and exploring more sophisticated feature extraction techniques.