Hierarchical Bayesian modeling of multi-country mortality rates

As world populations age along with speedy internationalization, forecasting mortality for multiple countries or populations with similar socio-economic conditions or close cultural connections has become essential. We apply the hierarchical Bayesian theory to the random walk with drift model governing the dynamics of the logarithm of central death rates for each age and population. Using the mortality data for both genders of three developed countries for an age span 25–84 and a series of fitting age-year windows, and further extending the data set to include both genders of twenty countries, we conclude that the proposed hierarchical Bayesian framework can more accurately capture the mortality trends and overall outperforms the Lee–Carter model and its three extensions in mortality forecasting. Grouping both genders of the twenty countries in three ways, we find that the expected improvement rates per year in the logarithm of central death rate for all twenty countries converge to about 2% except for the US.