Here's a bold statement: The way we predict credit card delinquency is about to change dramatically. But here's where it gets controversial—what if combining credit and debit data could not only improve prediction accuracy but also reveal deeper insights into why people fall behind on payments? A groundbreaking study from researchers at BI Norwegian Business School and NHH Norwegian School of Economics suggests exactly that. Published in The Journal of Finance and Data Science (https://doi.org/10.1016/j.jfds.2025.100166), this research introduces a hierarchical Bayesian behavioral model that outperforms cutting-edge machine learning algorithms like XGBoost, GBM, and neural networks in predicting credit card delinquency.
The key? Integrating debit transactions with credit card data. And this is the part most people miss—debit data provides a window into payday spending habits, repayment behavior, and income patterns, which are critical in understanding financial risk. As lead researcher Håvard Huse explains, "Credit data alone only tells half the story. By adding debit transactions, we can see how people manage their money day-to-day, which is essential for predicting delinquency."
Using data from a large Norwegian bank, the team found that traditional models, which rely on monthly aggregates like balance and credit limit, fail to capture the behavioral dynamics that drive repayment problems. Their new model, however, tracks how spending and repayment patterns evolve over time, offering a clearer picture of who is at risk and why. For instance, it identifies distinct behavioral segments, such as customers in financial distress who are more influenced by past financial states—a nuance standard models often overlook.
Here’s the controversial part: While the model is more accurate, it’s also more interpretable, which challenges the notion that complexity equals effectiveness. As co-author Auke Hunneman points out, "Banks don’t just need predictions—they need to understand the 'why' behind the risk."
The practical implications are significant. With a three-month prediction horizon, banks could intervene early, potentially saving costs and helping customers avoid financial crises. As co-author Sven A. Haugland notes, "This isn’t just about improving accuracy—it’s about proactively supporting customers."
This research marks a shift from static credit scoring models to dynamic behavioral analytics, raising a thought-provoking question: Are we ready to embrace a more holistic approach to credit risk assessment? Let’s discuss—do you think integrating debit data is the future of credit scoring, or does it raise privacy concerns? Share your thoughts in the comments!
