Credit risk prediction for small and medium enterprises utilizing adjacent enterprise data and a relational graph attention network

Jiaxing Wang, Guoquan Liu*, Xiaobo Xu, Xinjie Xing

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

Credit risk prediction for small and medium enterprises (SMEs) has long posed a complex research challenge. Traditional approaches have primarily focused on enterprise-specific variables, but these models often prove inadequate when applied to SMEs with incomplete data. In this innovative study, we push the theoretical boundaries by leveraging data from adjacent enterprises to address the issue of data deficiency. Our strategy involves constructing an intricate network that interconnects enterprises based on shared managerial teams and business interactions. Within this network, we propose a novel relational graph attention network (RGAT) algorithm capable of capturing the inherent complexity in its topological information. By doing so, our model enhances financial service providers' ability to predict credit risk even in the face of incomplete data from target SMEs. Empirical experiments conducted using China's SMEs highlight the predictive proficiency and potential economic benefits of our proposed model. Our approach offers a comprehensive and nuanced perspective on credit risk while demonstrating the advantages of incorporating network-wide data in credit risk prediction.

Original languageEnglish
Pages (from-to)177-192
Number of pages16
JournalJournal of Management Science and Engineering
Volume9
Issue number2
DOIs
Publication statusPublished - Jun 2024

Keywords

  • Credit risk
  • Graph neural network
  • Risk management
  • SME
  • Transactional data

Fingerprint

Dive into the research topics of 'Credit risk prediction for small and medium enterprises utilizing adjacent enterprise data and a relational graph attention network'. Together they form a unique fingerprint.

Cite this