From text to risk: Predicting repayment risk in supply chain finance with deep learning and large language models

Supply chain finance (SCF) plays a pivotal role in enhancing liquidity and operational efficiency for supply chain enterprises, particularly for capital-constrained suppliers. However, repayment risk—the failure of borrowers to meet payment obligations—can propagate through supply networks, causing systemic financial disruptions and production delays. While existing studies predominantly rely on structured financial data (e.g., financial statements) and periodic reports for risk assessment, these approaches suffer from timeliness gaps and fail to reflect real-time operational vulnerabilities. To address these limitations, this study proposes DeepRRP, a novel repayment risk prediction (RRP) framework that integrates the latest disclosures (e.g., litigation announcements, executive changes) with conventional financial indicators. Leveraging advanced large language models (LLMs) and deep learning algorithms, DeepRRP captures risk signals from disclosure texts and models their temporal dependencies, enabling proactive risk mitigation. Empirical tests on Chinese listed firms demonstrate that textual features from the latest disclosures can effectively compensate for the timeliness lag of traditional financial data and improve the accuracy of RRP. In addition, interpretable analysis identifies key factors affecting repayment risk, helping practitioners optimize lending terms and suppliers manage working capital.