Modelling and Forecasting Mortality Patterns

his work considers the problem of modelling and forecasting mortality. In the context of the Lee-Carter model, where the overall temporal pattern of mortality is well described by a time-varying index moderated by age-specific effects, the study proposes a sequential testing procedure valid in the presence of a structural change in the index of mortality, which allows the identification of the appropriate model used in forecasting future mortality patterns. Testing procedures proposed are applied to Portugal, and other eighteen developed countries, using data for the period 1950-2007. Structural changes in the overall time trend are found in Portugal, as well as in other developed countries, occurring mainly in male mortality. The impacts of a neglected structural change in the overall temporal behaviour of mortality are illustrated for the case of male Portuguese population. In order to better understand male and female mortality patterns in Portugal, several descriptive measures and visualization techniques are used and various extensions of the Lee-Carter model are applied. Changing rhythms of decline over ages and time are found. The possibility of a cohort effect in male mortality is also suggested. Finally, cohort specific influences in male Portuguese mortality patterns are studied using log-linear additive age-period-cohort models, as well as models allowing for age interaction with period and cohort effects. The evidence is favourable to the existence of cohort effects influencing male population patterns, with some generations experiencing stabilizing or even poor mortality conditions than preceding and subsequent cohorts.