Online Archive of University of Virginia Scholarship
Behavioral Credit Risk Modeling for Buy Now, Pay Later Systems: Evaluating Feature Design and Fairness Trade-offs Using Alternative Data; Constructing Trustworthiness: An Actor-Network Theory Analysis of Klarna’s Behavioral Credit Risk Models in Buy Now, Pay Later Lending 8 views
Author
Syed, Altaf, School of Engineering and Applied Science, University of Virginia
Advisors
Elliott, Travis, AT-Academic Affairs, University of Virginia
Morrison, Briana, EN-Comp Science Dept, University of Virginia
Abstract
Buy Now, Pay Later (BNPL) platforms increasingly rely on behavioral data rather than traditional credit scores to make lending decisions, raising fairness concerns for consumers with irregular income. I developed a simplified credit-risk model using synthetic borrower data to examine how alternative data features influence approval outcomes. Using logistic regression and decision-tree classifiers evaluated under both unconstrained optimization and fairness-aware constraints, the model produced approval gaps of 48–55 percentage points between regular- and irregular-income borrowers despite comparable true default rates. Fairness constraints reduced these gaps to under 4 percentage points with modest accuracy trade-offs. These findings highlight how technical feature design decisions can embed assumptions about trustworthiness into automated credit systems.
Syed, Altaf. Behavioral Credit Risk Modeling for Buy Now, Pay Later Systems: Evaluating Feature Design and Fairness Trade-offs Using Alternative Data; Constructing Trustworthiness: An Actor-Network Theory Analysis of Klarna’s Behavioral Credit Risk Models in Buy Now, Pay Later Lending . University of Virginia, School of Engineering and Applied Science, BS (Bachelor of Science), 2026-05-07, https://doi.org/10.18130/ear5-sp59.
