Complementing the evaluation toolkit of mortality forecasts with measures of lifespan disparity

Christina Bohk-Ewald, Max-Planck Institute for Demographic Research
Marcus Ebeling, Universität Rostock
Roland Rau, Universität Rostock

Evaluating the predictive ability of mortality forecasts is important and yet, at the same time, difficult. Average lifespan and death rates are basic life table functions that are typically used to analyze how much forecasts deviate from their realized values. While these parameters are useful to specify how precisely mortality has been forecasted at a certain point in time, they cannot be used to indicate whether the underlying mortality developments are plausible, too. We therefore propose to look in addition at lifespan disparity to examine whether the forecasted variability of the age at death is a plausible continuation of past trends. Validating mortality forecasts for Italy, Japan, and Denmark demonstrate that their predictive performance can be evaluated more comprehensively when analyzing average lifespan and lifespan disparity at the same time, i.e. jointly analyzing mean and dispersion of mortality. Approaches accounting for dynamic age-shifts in survival improvements outperform others that enforce invariant patterns.