Consumer Credit Risk in Embedded Finance: A Comparative Analysis of Traditional Scoring Methods and Machine Learning Models

Abstract

Credit risk assessment is critical in financial services, influencing decisions on loan approvals and interest rates. The objective of this study is to understand the impact of instalment payments on a client’s credit risk, particularly as a proxy for assessing the risk associated with Buy Now, Pay Later (BNPL) services as well as comparing distinct methods of credit risk assessment to conclude which one would work better in a world of digital financial services. Data from clients with and without instalment payments was analysed and compared, using both a traditional Scorecard model and a machine learning-based XGBoost model. The analysis revealed that clients with instalment payments typically displayed a higher credit risk, with XGBoost demonstrating superior performance across most performance metrics as well as distinguishing between clients' risk profiles. Accurate credit risk models allow for better credit decisions, potentially reducing default rates and contributing to financial stability, while providing insights into the risks associated with BNPL services, in this case. This research highlights the impact of embedded finance services on credit risk and the need for updated methods of credit scoring that can adapt to the fluid characteristics of new financial products, such as BNPL. This project allows for drawing conclusions about the impact of BNPL services on credit risk both in general and within the Portuguese context.