City Research Online

Abstract

A complete framework for the statistical modelling of excess mortality within the actuarial context is described, based on the theory of generalised linear models. In this context, the measure of excess mortality considered is the standardised mortality ratio. The modelling framework allows model building using several explanatory factors. The statistical significance of explanatory factors can be tested and, furthermore, the effect of covariate interactions can be assessed. Residual analysis is considered as a means of model checking and allows systematic and isolated departures from the model to be identified. A convenient and practically expedient method of calculating model based confidence intervals for the mortality ratio is developed. Three particular model structures are considered (the multiplicative, additive and power structures) and a unified approach to modelling excess mortality is presented.

The modelling approach has appealing connections with the traditional actuarial approach to the measurement of excess mortality and to the numerical rating system used almost universally in life insurance underwriting. These connections are explored and it is proposed that the modelling approach offers a scientifically sound approach to life insurance underwriting.

The modelling approach is used to analyse an extensive data set, namely the Prudential Impaired Lives data set, and the results are compared with previous results based on that data set and also with underwriting manuals in current use.

Publication Type:	Thesis (Doctoral)
Subjects:	H Social Sciences > HA Statistics
Departments:	Bayes Business School > Actuarial Science & Insurance Bayes Business School > Bayes Business School Doctoral Theses Doctoral Theses

Export

Downloads