The projection of future mortality experience constitutes a challenge for both actuaries and demographers. As we show, some of the currently used standard mortality projections have several shortcomings which may pose a serious threat to insurers, pension funds, and social security systems.

In this paper, we propose a new projection methodology which overcomes these shortcomings. We introduce a model which allows mortality improvements to depend on age, period, and cohort and we explain how the model can be estimated and applied. In particular, we show how coherent projections for several populations, i.e. males and females of the same country and populations from closely related countries, can be derived. The basis for these projections are coherent extrapolations of historical life expectancies. As aggregated mortality statistics, life expectancies typically exhibit steady patterns which often makes forecasting rather obvious. We observe that the incorporation of information on the mortality experience of other populations can have a significant impact on the projection for a given population. A comparison with other commonly used projection models shows that our model provides stable and highly plausible projections. Finally, we discuss uncertainties in our modeling approach and explain how they can be accounted for. In order to illustrate our methodology, we derive fully specified projections for German males and females as members of a large reference set of European populations.

*Awarded PBSS Track Prize