Statistical Challenges in Retail Credit Analysis
The retail credit domain is characterised by data sets which are large in terms of number of cases, number of variables, and acquisition rate. Furthermore, the area presents many novel statistical and mathematical challenges, requiring the development of new methods. This paper outlines some of the areas in which the Consumer Credit Research Group has contributed to the industry over recent years, including developing new measures of loan application scorecard performance, tools for detecting fraudulent credit card transactions, and methods for tackling selection bias in fraud and other areas.
KeywordsRandom Forest Credit Card Gini Coefficient Score Distribution Kolmogorov Smirnov
Acknowledgement of sponsors
Many bodies have sponsored the work of the Consumer Credit Research Group, including the EPSRC, ESRC, GMAC, HBOS, British Credit Trust, Capital One, Fair Isaac, Goldman Sachs, Barclaycard, Littlewoods, Barclays Direct Loan Division, Abbey National, Institute of Actuaries, Link Financial, Shell, and others. We are most grateful to all of them, for their vision and encouragement during our research.