Predicting High-Spending Customers Using Semiparametric Quantile Regression

• Activity Number: 667
• Type: Contributed
• Date/Time: Thursday, August 4, 2016 : 8:30 AM to 10:20 AM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Adam Maidman* and Lan Wang
• Companies: University of Minnesota and University of Minnesota
• Keywords: Expenditure prediction ; Lower tails ; Partially linear additive model ; Quantile regression ; Semiparametric regression ; Upper tails

Abstract

Motivated by a credit card expenditure data set, we consider a new semi- parametric approach for predicting if a credit card applicant is likely to be a high-spending customer in the sense that his/her future spending will occur in the upper tail of the expenditure distribution. The proposed method is useful for a general class of important applications that involve predicting whether a future response occurs at the upper/lower tail of the response distribution. The new method enjoys some nice features: (1)It does not require an artificially dichotomized response and hence better uses the information contained in the data; (2)It does not require any parametric distributional assumptions and hence tends to be more robust; (3)It incorporates nonlinear effects of the covariates; and (4)It can be easily adapted to construct a prediction interval and hence provides more information about the future response. We show that with probability approaching one, the new method makes the correct prediction on whether the future response exceeds a given threshold. We make available an R package to implement the procedure and illustrate the application of the new method on the credit card data.