Hybrid Seminar Talk
Lee-Carter Cohort Mortality Forecasts
Laboratory of Digital and Computational Demography, May 12, 2022
The seminal model of Lee and Carter (LC) has stimulated great advances in the field of mortality forecasting since its introduction in 1992. However, applications of the model to forecast cohort mortality are still lacking in the literature. In this talk, we start by showing the shortcomings of the “state-of-the-art” approach of forecasting period mortality and extracting cohort patterns from the Lexis diagonals. Next, we adapt the estimation of the LC parameters to the structure of cohort mortality data. Mortality forecasts of partially observed cohorts can be directly derived from the estimation of the LC parameters. Nonetheless, we show that LC forecasts of cohort mortality may be inaccurate and biased. We therefore propose the “Linear Lee-Carter” (LLC) model, a special case of the LC where the cohort index kc is modelled as a linear function of cohorts. Particular care is given to smooth LLC forecast age-patterns while reducing the jump-off bias. We apply the LC and LLC models to cohort mortality data at ages 0-100+ for the cohorts 1900-1990 of two populations obtained from the Human Mortality Database. We show that LLC forecasts are more accurate and reasonable than those of the other two approaches. Our proposed LLC model is a powerful approach for forecasting mortality of partially observed cohorts.
Ugofilippo Basellini is a research scientist in the Laboratory of Digital and Computational Demography at the Max Planck Institute for Demographic Research, and an associated researcher at the Institut national d’études démographiques. His main research interests are related to statistical demography, with a particular focus on mortality modelling and forecasting, lifespan inequality and formal demographic methods for the study of mortality.