A1317
Title: Threshold networks in credit risk models: An application on P2P markets
Authors: Eduard Baumohl - Masaryk University (Czech Republic) [presenting]
Stefan Lyocsa - Slovak Academy of Sciences (Slovakia)
Tomas Vyrost - Technical university in Kosice (Slovakia)
Abstract: P2P lending markets offer risky investment opportunities where accurate credit risk models are in high demand. Publicly available loan books might offer a broad spectrum of loan and borrowers' characteristics that lead to high-dimensional systems that make the usage of traditional credit scoring models challenging. The purpose is to explore whether complex relationships between risky assets (loans) can be identified via $\alpha\%$ threshold feature-based networks. More specifically, adjacency matrices are created using heterogeneous distance measures, and networks are built with the $\alpha\%$ threshold approach. Topological properties are extracted as loan-based features to augment credit risk models. A statistical comparison uncovers the most promising network-based features for improving credit risk models.
