Graph Neural Networks for Interconnected Default Risk in Supply Chain Finance Portfolios

Abstract

In supply chain finance, credit risk is not isolated defaults propagate across interconnected firms due to shared financial dependencies, contractual obligations, and operational linkages. Traditional credit risk models often treat obligors independently, ignoring the complex network structures that define modern supply chains. In this paper, we propose a novel approach using Graph Neural Networks (GNNs) to model interconnected default risk within supply chain finance portfolios. By representing firms and their trade relationships as nodes and edges in a graph, we exploit the expressive power of GNNs to learn both firm-specific credit characteristics and inter-firm risk contagion dynamics. We construct synthetic and real-world supply chain datasets to evaluate our framework, comparing it against traditional risk scoring and machine learning baselines. Our results show that the GNN-based approach significantly outperforms non-graph methods in predicting defaults, especially in scenarios involving cascading failures or sectoral stress. We also explore explainability tools tailored for GNNs to provide interpretable insights into systemic risk sources. This work highlights the value of graph-based modeling in understanding and mitigating interconnected risks in financial ecosystems.