A HGLM Approach for Clustered Insurance Claim Data

• 발행기관: 한국자료분석학회
• 저자명: Il Do Ha Maengseok Noh
• 간행물 정보: 『Journal of The Korean Data Analysis Society (JKDAS)』Vol.11 No.6, 2893~2904쪽, 전체 12쪽
• 주제분류: 자연과학 > 통계학
• 파일형태: PDF
• 발행일자: 2009.12.30

영문 초록

Various insurance data with independent responses have been widely analyzed using generalized linear models(GLMs). Recently, the analyses of clustered insurance data which are usually clustered by policies, regions, divisions or individuals have been studied. Such data often have not only a correlation within clusters but also a heterogeneity between clusters. Thus the data can be analyzed by using extended GLMs with random effects. However, their usual inference methods require intractable integrations and don’t provide a proper prediction for random effects. In this paper we propose a new framework for analyzing the clustered claim data using hierarchical GLMs(HGLMs. Lee and Nelder, 1996). For the inferences we use the hierarchical likelihood(h-likelihood) which avoids such difficult integrations and provides a statistically efficient procedure. Furthermore, we also propose how to construct the prediction intervals of random effects which are very useful in investigating a potential heterogeneity across clusters. The proposed method is illustrated with a real data set about the number of third party claims which are clustered by 13 divisions in Australia.

목차

1. Introduction

2. Example: Third-party claim data

3. Model Estimation

4. Analysis

5. Conclusion

Reference

키워드

Clustered claim data Hierarchical generalized linear models H-likelihood Prediction interval Random effects