Some Empirical Evidence on the Need of More Advanced Approaches in Mortality Modeling

Recent literature on mortality modeling suggests to include in the dynamics of mortality rates the effect of time, age, the interaction of the latter two terms and finally a term for possible shocks that introduce additional uncertainty. We consider for our analysis models that use Legendre polynomials, for the inclusion of age and cohort effects, and investigate the dynamics of the residuals that we get from fitted models. Obviously, we expect the effect of shocks to be included in the residual term of the basic model. The main finding here is that there is persistence in the residual term but the autocorrelation structure does not display a negative exponential behavior. This empirical result suggests that the inclusion of the additional shock term requires an appropriate model that displays a more flexible autocorrelation structure than the Ornstein-Uhlenbeck employed in existing models.