The Relationship Between Default Prediction and Lending Profits: Integrating ROC analysis and loan pricing
by Roger M. Stein of Moody's KMV
Abstract: In evaluating credit risk models, it is common to use metrics such as power curves and their associated statistics. However, power curves are not necessarily easily linked intuitively to common lending practices. Bankers often request a specific rule for defining a cut-off above which credit will be granted and below which it will be denied. In this paper we provide some quantitative insight into how such cut-offs can be developed. This framework accommodates real-world complications (e.g., ''relationship" clients). We show that the simple cut-off approach can be extended to a more complete pricing approach that is more flexible and more profitable. We demonstrate that in general more powerful models are more profitable than weaker ones and we provide a simulation example. We also report results of another study that conservatively concludes a mid-sized bank might generate additional profits on the order of about $4.8 million per year after adopting a moderately more powerful model.
Keywords: Default prediction, ROC analysis, Model validation, Rating systems, Economic value of information, Discrete choice models, Model power, Simulation, Bank lending.
Published in: Journal of Banking & Finance, Vol. 29, No. 5, (May 2005), pp. 1213-1236.
Previously titled: Power, Profitability, and Prices: Why Powerful Models Increase Profits and How to Define A Lending Cutoff If You Must