Predictive Power of Stochastic Mortality Models for Different Age Groups: An Analysis of Developing and Developed Countries

Abstract

Mortality modelling helps in evaluating longevity and mortality risks by actuaries and demographers. The performance of mortality models strongly depends on the different salient features observed in different countries. In this paper, a comparative study of the Lee-Carter (LC) model, Age-Period-Cohort (APC) model and Cairns-Blake-Dowd (CBD) stochastic mortality models is presented for different age groups (0 to 4 years, 5 to 60 years and above 60 years) from developed and developing countries. The developed countries are France, Japan and Canada, and the developing countries are Ukraine and Taiwan. The forecast accuracy are evaluated using RMSE, MAE, MPE and MAPE metrics. Results indicate that models that are best for some age groups for developing countries might not be appropriate for developed ones since the evolution of each country’s mortality rates is different from one another. Yet, for some ages, the same mortality model can be used to project mortality forecasts for both developing and developed countries. For higher ages (75+), the CBD model is found to outperform the other models across all countries.