Modeling Edge-Level Risk Propagation for Customer Scoring in Cheque Factoring Networks

Factoring companies face significant challenges in assessing credit risk when accepting cheques with complex obligator chains. Traditional credit scoring focuses solely on drawers, ignoring network effects and customer side risk propagation. We propose a novel graph-based analytical framework that scores customers based on their position within payment networks. Our approach models risk propagation by calculating a composite risk score for every edge in a customer's network map, combining the child node's intrinsic risk (GDR, credit score) with the edge's financial exposure (utilization, volume). These edge risks are aggregated with a distance-decay function to produce a final customer score. Through experiments on 21 months of real-world data comprising 47,382 customers, we demonstrate that our method achieves $\mathbf{8 0 \%}$ F1-score in predicting defaults, outperforming traditional drawer-focused scoring (61% F1) and ML baselines (71% F1). Critically, our interpretable formula avoids complex model fitting and requires only minimal calibration, enabling immediate deployment. This work is the first to show that a customer's network topology, when modeled at the edge-level, can rival ML performance while maintaining full explainability.