Machine learning modeling for credit card risk management and profit optimization
![Quantitative Finance Lecture Series@MIDS Quantitative Finance Lecture Series@MIDS](/images/2023/20231114/f2b5ad72c90e5ad4e4b2b05bbe590e6d-CR-MIDS QF general slide_20231110055539PM.png)
We start the talk with the fundamentals of modeling practices used in retail credit decisioning. This includes data, how to process it, and typical modeling methodologies. We then discuss model risk management fundamentals such as topics in performance monitoring and interpretability analysis.
After this foundational talk, we delve into the use cases of these models in predicting risk or profit for credit card acquisitions strategies and their implications for credit risk. First, we explore the relationship of empirical risk and profit in credit card portfolios. Then, we examine the implications of switching from a traditional risk-based underwriting approach using logistic modeling to a profit-based underwriting approach using machine learning models. We conclude by comparing simulated risk vs. profit-ranked portfolios in terms of customer profiles, showcasing the importance of model methodology in retail banking.
George Krivorotov is a senior financial economist in the Retail Credit Risk Analysis Division at the Office of the Comptroller of the Currency within the US. Treasury. George's research interests include climate risk, machine learning, real estate economics, and quantitative macroeconomics. At the OCC, George provides technical consultation and direct supervision support on a wide variety of topics, focused on complex models used in credit allocation, loss forecasting, climate risk management, and collateral valuation. He holds a doctorate in economics from the University of Minnesota and an undergraduate degree in mathematics from the University of Virginia.