Social media-driven credit scoring: The predictive value of social structures

Abstract

While emerging economies have seen an explosion of social network site (SNS) adoption, these countries lack sophisticated credit scoring system or credit bureaus to predict creditworthiness of individuals. In this paper, we propose an SNS-based credit scoring method for micro loans using largescale observational data. We show empirical evidence that by incorporating social network metrics, we can improve the repayment prediction by 18%. Next, to better harness the combination of borrowers and non-borrowers, we implement graph-based prediction using the semi-supervised learning method. At current stage, by only leveraging on social ties, the prediction performance looks promising with an accuracy of 60%. We believe that although lending to the poor without incurring high default rates is challenging, the SNS-based methods can be effective for developing countries that face the "cold-start" problem.