Multi-Population Mortality Modelling with L evy Processes

Abstract

This paper constructs a theoretical framework for multi-population mortality modelling by introducing generalized linear models and Levy stochastic perturbations driven by a common and an idiosyncratic factor. To accommodate the evolution of factor loadings across age and calendar time, we specify the loading coefficients on common factors to be deterministic non-linear functions of age and calendar time. Different model formulations are explored by using Levy processes such as Gamma process and Variance Gamma process, and their in-sample fitting and out-of-sample forecasting performances are evaluated. In addition, we systematically investigate the dependence structure of both the inter-population and cross-population kind. In our empirical investigations, the mortality experiences of male and female lives in UK over the period 1964-2013 are used.