Reproduce “A Multimodal Approach to SME Credit Scoring Integrating Transaction and Ownership Networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sahab Zandi, Kamesh Korangi, Juan C. Moreno-Paredes, María Óskarsdóttir, Christophe Mues, Cristián Bravo
The study from Western University, University of Southampton, and Reykjavík University introduces a multimodal framework that integrates Graph Neural Networks with multilayer transaction and ownership networks, alongside traditional financial data, to enhance Small and Medium-sized Enterprise credit risk prediction. This approach achieved an AUC of 0.879 and an AUCPR of 0.414, surpassing traditional models, and provides insights into risk propagation mechanisms.
Key insights
- Multilayer networks, particularly financial transaction data, offer more informative representations for SME credit risk assessment than single-layer or static common ownership data.
- Incorporating edge directionality and weights reveals that buyer-side defaults (customers) have a greater adverse impact on an SME's default probability than supplier-side defaults.
- The proposed model effectively leverages both network topology and traditional financial features, with network data showing a slightly higher average contribution to the final prediction.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.