Bankruptcy prediction for SMEs: The role of relational network and information propagation
提出图神经网络模型HG-HAN，利用非财务信息和公司间关系预测中小企业破产，发现网络变量通过信息传播提升预测效果，对金融机构和监管者改进预警系统有实际意义。
Predicting bankruptcy for small- and medium-sized enterprises (SMEs) is important for stabilizing supply chains, reducing bank loan risks, and mitigating financial risks. To address the challenge of obtaining comprehensive financial data from SMEs, we propose a graph neural network (GNN) model, HG-HAN, which utilizes nonfinancial information and interfirm relationships. We innovatively combine macroscopic network centrality indicators with microscopic information propagation in GNN, enabling a deeper exploration of network structure and risk contagion. Experiments on real-world data sets demonstrated that network variables enhance prediction by capturing the propagation of information through relational networks, especially transactional, investment, and industry relationships. We also found that SMEs embedded in cohesive and well-connected clusters have greater informational advantages and a lower probability of failure. These insights provide practical implications for financial institutions and regulators in improving early-warning systems and contribute to SME risk management theory by highlighting the role of interfirm relationship networks in bankruptcy.