Statistical challenges in credit card issuing (Q2722294)

The large-scale delivery of customer credits, whether it be for instalment loans, mortgages or credit cards, would not be possible without the use of various statistical techniques for evaluating good and bad credit risks. The primary technique in use is credit scoring and its derivative, behaviour scoring. Credit scoring originated during World War II with \textit{Henry Wells}, an executive of Spiegel Inc.. However, it only really started to come into prominence in the late 1950s and early 1960s when \textit{Bill Fair} and \textit{Earl Isaac}, two young Americans, realized that statistical discriminant analysis could assist in predicting whether customers would default on their credit agreements. They then established the company ``Fair Isaac Inc.'' (one of the most eminent credit scoring companies in the world), and successfully used computers in the scorecard construction process. Credit scoring was transported to the U.K. in the 1970s. Behaviour scoring was a later development based upon the same principles, but whereas credit scoring evaluated whether a new customer was too risky to be given credit, behaviour scoring evaluated the probability of existing customers defaulting. The company where the author works (Barclaycard, Northampton, U. K.) has over 7 million customers and over 9 million credit cards in circulation and deals with over 300 million credit card transactions per annum -- a tremendous amount of data. Hence, the credit card industry is awakening to the fact that the vast amounts of data it has on its customers should be treated as a valuable asset and that extracting knowledge from this data is crucial if one wants to optimize ones credit book.NEWLINENEWLINENEWLINEThis paper is intended to provide a flavour of some of the statistical work that is required to take advantage of this asset. The traditional statistical methodology behind the issues of credit, along with some of the problems that have been solved and some that are still outstanding is addressed. The paper also provides an overview of the new Customer Value Management paradigm, also known as Customer Relationship Management, highlighting the new challenges that have emanated from this and the implications for statisticians.