ITEM METADATA RECORD
Title: Using mixture cure models with unobserved heterogeneity for the analysis of credit loan data
Authors: Dirick, Lore
Claeskens, Gerda
Vasnev, A.
Baesens, Bart
Issue Date: 2014
Conference: International Conference on Computational Statistics (COMPSTAT 2014) edition:21 location:Geneva (Switzerland) date:19-22 August 2014
Abstract: Due to more strict regulations as a result of the Basel accords, survival analysis is becoming more popular in the field of credit risk analysis due to a large number of censored cases. As an extension to survival models, multiple event mixture cure models are used in order to model several event types for a credit loan (default, early repayment and maturity) jointly in one model. In our research, the multiple event mixture cure model is extended by allowing for unobserved heterogeneity within the event groups. This way, different parameter estimates are possible for different subject groups. A hierarchical EM-
algorithm is used to model the (higher level) event types on one hand, and the unobserved heterogeneity on the other, resulting in parameter estimates through the maximization of the expected complete-data log likelihood. We perform a simulation study in which it is shown that allowing for heterogeneity can result in improved prediction accuracy compared to the multiple event mixture cure model without
unobserved heterogeneity. Additionally, using a real life credit data example, we illustrate that there is indeed heterogeneity present in the group of subjects that repay their loans early.
Publication status: published
KU Leuven publication type: IMa
Appears in Collections:Research Center for Operations Research and Business Statistics (ORSTAT), Leuven
Research Center for Management Informatics (LIRIS), Leuven

Files in This Item:

There are no files associated with this item.

 


All items in Lirias are protected by copyright, with all rights reserved.