"A Markov Multinomial Regression Model for Predicting Consumer Credit R" by Thomas Brandenburger

Document Type

Thesis - University Access Only

Award Date

2009

Degree Name

Doctor of Philosophy (PhD)

Department / School

Mathematics and Statistics

Abstract

This paper proposes a Markov model for use in the prediction of credit failure of credit card holders, where the transition probabilities are modeled through a multinomial regression with predictors of risk as covariates. Examples using actual industry data will be given. In order to measure the performance of the model a new Model Quality Metric is proposed as an improvement to the commonly used Kolmogorov-Smirnov Statistic and Receiver Operating Characteristic Curve.

Library of Congress Subject Headings

Consumer credit
Credit cards
Risk management -- Mathematical models
Markov processes

Format

application/pdf

Number of Pages

135

Publisher

South Dakota State University

Share

COinS