Lee-Carter Cohort Mortality Forecasts

Résumé 0

Mortality forecasting has recently stimulated great interest in academics and financial sector practitioners due to the increasing challenges posed by continuous longevity improvements. Many efforts have been directed to model and forecast mortality from the conventional age-period perspective. Conversely, models to forecast cohort mortality are scarce in the demographic and actuarial literature. Here, we fill this gap by proposing a modification of the seminal model of Lee and Carter (1992) to account for the structure of cohort mortality data. Particular care is given to smooth the forecast age-pattern of mortality while reducing the jump-off bias. Our proposed "cohort-LC" model allows us to derive a complete age-cohort mortality surface simply by estimating the LC parameters, thereby completing the mortality experience of non-extinct cohorts without performing time-series forecasting. We apply our methodology to cohort mortality data at ages 0-100 for the cohorts 1900-1990 obtained from the Human Mortality Database.