Delinquent Events Prediction in Temporal Networked-Guarantee Loans.

Area of Science:

• Financial Risk Management
• Network Science
• Machine Learning

Background:

• Small and medium-sized enterprises (SMEs) utilize mutual guarantees to secure bank loans, creating interconnected networks.
• Economic downturns can amplify default risks within these complex SME guarantee networks, potentially leading to systemic financial crises.
• Traditional loan assessment methods struggle to capture temporal dynamics and structural positions influencing SME default probabilities.

Purpose of the Study:

• To develop a novel framework for predicting SME loan defaults by incorporating temporal network structures and credit behavior.
• To enhance macroprudential oversight and mitigate systemic financial risk within SME guarantee networks.
• To improve the accuracy and efficiency of loan risk assessment for commercial banks and regulatory bodies.

Main Methods:

• Proposed a temporal delinquent event prediction (TDEP) framework using an end-to-end model.
• Employed a graph attention layer to learn node representations in dynamic guarantee networks.
• Designed a recursive and self-attention mechanism to integrate credit behavior and network structure data.

Main Results:

• The TDEP framework effectively preserves temporal network structures and credit behavior sequences.
• Attentional weights identified high-risk guarantee patterns, accelerating risk assessment.
• Extensive experiments on a real-world dataset demonstrated superior performance compared to state-of-the-art baselines.

Conclusions:

• The proposed TDEP framework offers a robust solution for predicting SME loan defaults in dynamic network environments.
• Integrating temporal and structural network information significantly enhances risk assessment accuracy.
• The model's successful integration into a real-world loan risk management system validates its practical applicability and effectiveness.