The value of detailed product information in credit risk prediction : A case study applied to Klarna’s Pay Later orders in Sweden

Abstract: In this study we propose to enhance the predictive power of a Buy Now, Pay Later (BNPL) consumer credit scorecard by leveraging detailed product information. The object of analys is in this study is Klarna Bank AB, which is the largest retail finance provider in Sweden. This research conducts a quantitative study in order to firstly, investigate if it is possible to find subcategories that correlate more with credit risk than the existing product categories at Klarna. This will be investigated by categorizing already accepted orders into more granulated product categories than Klarna's existing level. Secondly, this study investigates how more detailed product categorization can improve a BNPL e-commerce consumer credit scorecard. Lastly, a qualitative analysis of what the business impact an implementation of this feature could entail for Klarna Bank AB is conducted. Our results demonstrate that it is possible to find subcategories that correlate more strongly with credit risk than the existing categories. The characteristics of the high-and low risk product categories align with existing research on online consumer behavior. More specifically, we found that luxury products, ego-related products, and products related to addictive behavior had the highest risk. We also contribute to existing research within the credit risk management field by finding that trending/new products on the market have a higher risk, and that the novelty of a product should be taken into consideration in credit risk prediction. By applying a hypothetical credit scoring model on a dataset of already accepted orders that took the new detailed product categories into consideration, the discrimination performance could be improved. However, risks regarding adding more data into a credit risk model need to be considered before implementing the proposed solution. Our study, therefore, demonstrates the potential of including more granulated product category information in a BNPL e-commerce consumer credit scorecard to improve risk prediction. While the results of this study are limited to the studied context, it is considered generalizable in that the proposed method could effectively be adapted to retrieve corresponding findings in other contexts.