Modelling and forecasting mortality improvement rates with random effects (Q2066777)

The paper considers the forecast for mortality improvement rates over time. In this context, linear parametric structures for mortality improvement rates are considered. The aim of the study is to obtain a more unified approach, precisely considering the feasibility of modelling the main period effects as a random effect from the outset. After analyzing the inter-relationship between two alternative measures of mortality improvement rates, the paper presents a direct and indirect method of modelling mortality improvement rates, with a specific focus on the linear age-period-cohort structure. The importance of the inclusion of a random effects period component emerges from the analysis. Then, a fitting methodology for both normal-normal and Poisson-normal generalised linear mixed models is described, consistent with the specific framework of the study. Further deepening shows how a simple structured time series can be incorporated into the fitting process. The numerical investigation in the paper makes use of the recent UK male mortality experience.