Coherently forecasted mortality and its sensitivity to the method and the selection of countries

Lenny Stoeldraijer, Statistics Netherlands
Fanny Janssen, University of Groningen

In Western Europe, convergence has been observed in old-age mortality, and convergence in mortality between countries is likely thanks to common socio-economic policies, similar progress in medical technology, and shared importance of certain life style factors over time. However, mortality forecasts for single populations will in the long run lead to divergent outcomes, contrary to what would be expected based on historical trends. Therefore, different coherent forecasting methods are introduced, in which “coherent” refers to non-divergent forecasts for subpopulations within a larger population. A crucial question is the identification of the countries which determine the basic mortality trend that will be applied to other countries. The purpose of this research is to analyse the sensitivity of projected mortality based on different coherent forecasting methods and the sensitivity of the outcome due to the selection of the countries to take as a group. Data from several countries on all-cause mortality and population numbers by sex, age (0, 1-4, 5-9, …, 90-94, 95+), and year (1950-2009) will be obtained through the Human Mortality Database. The different coherent forecasting methods are the Li-Lee method and a coherent functional method. We will make out-of-sample forecasts up to 2050 to test the sensitivity of the method and choice of countries, and within-sample forecast to see how accurate the results will be. We expect that mortality forecasts for countries with historically a more linear mortality trend are more accurate, regardless which countries are taken as a group. For countries with less linear trends, we expect an important impact of the choice of the group of countries.