On the combination of graph data for assessing thin-file borrowers' creditworthiness

Abstract

The thin-file borrowers are customers for whom a creditworthiness assessment is uncertain due to their lack of credit history; many researchers have used borrowers' relationships and interactions networks in the form of graphs as an alternative data source to address this. Incorporating network data is traditionally made by hand-crafted feature engineering, and lately, the graph neural network has emerged as an alternative, but it still does not improve over the traditional method's performance. Here we introduce a framework to improve credit scoring models by blending several Graph Representation Learning methods: feature engineering, graph embeddings, and graph neural networks. In this approach, we stacked their outputs and trained a gradient boosting classifier to produce a single score. We validated this framework using a unique multi-source dataset that characterizes the relationships, interactions, and credit history for the entire population of a Latin American country, applying it to credit risk models, application and behaviour, targeting both individuals and companies. Our results show that the graph representation learning methods should be used as complements, and these should not be seen as self-sufficient methods as is currently done. In terms of AUC and KS, we enhance the creditworthiness assessment statistical performance, outperforming traditional methods to incorporate graph data. In Corporate lending, where the gain is much higher, it confirms that evaluating an unbanked company cannot solely consider its characteristics. The business ecosystem where these companies interact with their owners, suppliers, customers, and other companies provides novel knowledge that enables financial institutions to enhance their creditworthiness assessment. Our results let us know when and on which population to use graph data and what effects on performance to expect. They also show the enormous value of graph data on the unbanked credit scoring problem, principally to help companies' banking.

On the dynamics of credit history and social interaction features, and their impact on creditworthiness assessment performance

“…This study uses gradient boosted trees because they have consistently shown state-of-the-art performance in different problems (Chen and Guestrin, 2016;Friedman, 2001;Muñoz-Cancino et al, 2021). Additionally, in order to quantify the performance, we use the area under the receiver operating characteristic curve (Bradley, 1997, AUC) and the Kolmogorov-Smirnov statistic (Hodges, 1958, KS) as performance measures.…”

“…The data sources used to construct this dataset have already been used to develop credit scoring models (Muñoz-Cancino et al, 2021). Application and behavioral scoring models were trained.…”

“…Putra et al (2020) investigated the value of social network data to predict bankruptcy, and Letizia and Lillo (2019) included a corporate payments network to assess an internal rating. Our previous research (Muñoz-Cancino et al, 2021) shows that social network data generate a much more significant performance enhancement in application scoring than behavioral scoring, considering the same population and features. Moreover, this result is consistent when the credit scoring model is applied to both individuals and companies, personal scoring and business scoring, respectively.…”