Researchers from BI Norwegian Business School and NHH Norwegian School of Economics have developed a behavioral credit-risk model that integrates credit and debit transactions, significantly improving the prediction of credit card delinquency. The model outperforms state-of-the-art machine learning methods such as XGBoost, GBM, neural networks, and stacked ensembles, while also offering greater interpretability into the behavioral drivers of repayment problems.
The study, published in The Journal of Finance and Data Science, combines detailed credit and debit transaction data from a large Norwegian bank. Traditional credit-risk models rely heavily on monthly aggregates like balance and credit limit, which do not capture day-to-day financial management. The new model incorporates behavioral dynamics such as payday spending, repayment patterns, and income fluctuations, providing a more complete picture of a customer's financial situation.
"Credit data alone gives only a partial picture of a customer's financial situation," explained first author Håvard Huse. "By integrating debit transactions, we gain insight into payday spending, repayment behavior, and income patterns—factors that strongly influence whether someone is at risk of missing payments."
The hierarchical Bayesian model also identifies distinct behavioral segments with different "memory lengths," meaning the extent to which past financial states affect current repayment behavior. Customers in financial distress tend to be more influenced by earlier months' behavior, a dynamic that standard machine-learning tools fail to capture as effectively.
Using a three-month prediction horizon, the model enables early detection of at-risk cardholders, potentially generating substantial cost savings through timely intervention. "For banks, this is more than an accuracy improvement—it is a way to proactively help customers avoid serious financial problems," said co-author Sven A. Haugland.
The findings highlight a shift from traditional static credit scoring toward richer behavioral analytics based on a full picture of customer transactions. The research team included Håvard Huse (BI Norwegian Business School), Sven A. Haugland (NHH), and Auke Hunneman (BI). The study received no specific external funding. More information can be found at Chuanlink Innovations.
