Analysis of mortgage insurance data by mixture models

Liang Wang

Research output: Contribution to journalMeeting abstract

Abstract

Purpose: This paper demonstrates how mixture survival models can be applied to analyse mortgage insurance data that include default-prone and default-free loans, assess risk factors, and predict default rate. Originality: Although with proven advantages, mixture survival models have not previously been applied to mortgage insurance or other general insurance products with large numbers of default-free policies. Key literature / theoretical perspective: Mixture models have the flexibility of isolating default-free policies from the estimation of the survival function for the default-prone policies. Design/methodology/approach: We provide examples to identify and analyse the effects of two commonly used risk factors using the likelihood-ratio test and improper proportional hazard (PH) models. Moreover, given a set of plausible parametric models, we show how to select the best one based on the goodness of fit and model complexity. Findings: After applying both parametric and non-parametric estimation methods, we propose a Weibull mixture model for the survival function for default-prone policies. Research limitations/implications: The methodology applied in this research is ready to be extended to any other credit risk modelling. Practical and Social implications: Mortgage default is a crucial issue in assessing financial and insurance risks. It is well known that a large scale of mortgage defaults was the root of the sub-prime loan problems and the subsequent global financial crisis.
Original languageEnglish
Pages (from-to)87
Number of pages1
JournalExpo 2010 Higher Degree Research : book of abstracts
Publication statusPublished - 2010
EventHigher Degree Research Expo (6th : 2010) - Sydney
Duration: 19 Nov 201019 Nov 2010

Keywords

  • mortgage insurance
  • survival models
  • long-term survivors
  • Cox PH model

Fingerprint Dive into the research topics of 'Analysis of mortgage insurance data by mixture models'. Together they form a unique fingerprint.

Cite this