Factorial Network Models to Improve P2P Credit Risk Management.

Area of Science:

• Financial Technology (FinTech)
• Computational Finance
• Data Science

Background:

• Small and medium-sized enterprises (SMEs) face challenges in accessing credit, particularly within the peer-to-peer (P2P) lending landscape.
• Traditional statistical credit scoring models often struggle with the heterogeneity of SME populations, potentially leading to suboptimal risk assessment.
• Improving the accuracy of credit scoring models is crucial for both lenders and borrowers in the P2P lending market.

Purpose of the Study:

• To develop and evaluate an improved statistical credit scoring methodology for SMEs in P2P lending.
• To introduce a factor network-based segmentation approach for more accurate credit risk modeling.
• To compare the predictive performance of the proposed method against conventional logistic regression models.

Main Methods:

• Construction of a factor network representing SMEs based on the comovement of latent factors.
• Segmentation of the heterogeneous SME population into distinct clusters using the network structure.
• Development of cluster-specific credit score models employing lasso-type regularization logistic regression.

Main Results:

• The factor network-based segmentation effectively clusters heterogeneous SMEs.
• Credit score models built on these clusters demonstrate superior predictive performance.
• The proposed approach significantly outperforms conventional logistic models in credit risk assessment for P2P lending SMEs.

Conclusions:

• Network-based segmentation offers a robust framework for improving credit scoring accuracy in P2P lending.
• The methodology addresses SME population heterogeneity, leading to more precise credit risk evaluation.
• This advanced approach provides a valuable tool for financial institutions and P2P platforms operating in SME lending.