Document Type
Thesis - University Access Only
Award Date
2010
Degree Name
Master of Science (MS)
Department / School
Mathematics and Statistics
Abstract
This thesis studies reject inference in credit scoring. Credit models are often built on only those applications that have been accepted. This creates a bias issue since much of the target population may not be included in the model building process. Reject inference techniques attempt to make an educated guess on the probability the rejected applicants would have actually been a good observation. The data set analyzed is a sub-prime credit mailing in 2008 and 2009 from CAPITAL Card Services. Several reject inference techniques are discussed and extrapolation, stratification, and clustering were used to analyze the data. The specific models will be analyzed and compared using receiver operating characteristic charts, the Kolmogorov-Smirnov Statistic, and bias estimates. The goal of the paper is to discover if some techniques other than a random mailing can provide good predictive power while having a lower initial cost.
Library of Congress Subject Headings
Consumer credit
Credit analysis -- Mathematical models
Credit scoring systems -- Mathematical models
Format
application/pdf
Number of Pages
62
Publisher
South Dakota State University
Recommended Citation
Lutz, Nathaniel, "Reject Inference in Sub-Prime Credit" (2010). Electronic Theses and Dissertations. 1666.
https://openprairie.sdstate.edu/etd2/1666