Forecasting the Future: Using Lagged Data to Improve Coherent Mortality Forecasting

Heather Booth , Australian National University
Marie-Pier Bergeron-Boucher, Interdisciplinary Center on Population Dynamics (CPop), University of Southern Denmark

The use of a standard (or external reference population) in coherent mortality forecasting can improve the accuracy of the forecast, depending on the choice of standard. However, it is by no means clear how to choose a priori an external standard that will be advantageous. Previously used standards include the mortality of the other sex in sex-specific forecasting, the mortality of the national population in subnational forecasting, and the mortality of a group of populations. There is some evidence that a low-mortality standard is advantageous, acting as a guide in the mortality decline. This paper further develops the use of a low-mortality standard by using forward lagged data of the population of interest as the standard. Data for 21 countries for the period 1950-2014 are obtained from the Human Mortality Database. The forecasts are made using the product-ratio coherent method with functional data models (Hyndman, Booth and Yasmeen 2013 Demography). The two populations employed refer to the overlapping periods from 1950 to 2004 and to 2014, a lag of 10 years. The method produces forecasts commencing in 2015. Forecasts are evaluated against independent forecasts and sex-coherent forecasts, using the mean absolute relative error, MARE, and mean relative error, MRE, in age-specific mortality rates. Results show that forecast accuracy and bias are generally improved by using data lagged by 10 years, and that heterogeneity across countries is also reduced. The effect of lag length is also examined.

See extended abstract

 Presented in Session P2. Poster Session Ageing, Health and Mortality