Ahbab Mohammad Fazle Rabbi , University of Dhaka
Stefano Mazzuco , University of Padova
Heather Booth, Australian National University
Vladimir Canudas-Romo, Australian National University
Coherent mortality forecasting methods try to capture the influence of common improvement of health, communication, science on a specific population. The commonly used coherent method, Li-Lee (LL) method, is a hierarchical form of the Lee-Carter (LC) method. The LC model assumes an invariant age component and a presumably linear time component. The LL model fits the joint mortality data of populations which are believed to have closer interaction in terms of mortality regime (relational populations) and here the time component is adjusted to match observed life expectancy at birth. Choosing the appropriate reference population remains an arbitrary process. We propose to use smoothed mortality rates obtained by LASSO type regularization in LL model and hence to partially adjust the time component of the LL model according to observed lifespan disparity to get the common factor of the relational populations (reference group). Time variability is also taken into consideration during obtaining the common factor. The relational populations for making coherent forecast for a particular population is chosen from the set of available populations on the basis of closest lifespan disparity over time. The proposed methodology generates less forecast errors than existing method during the out-of-sample forecast period and also produce a more optimistic forecast of life expectancy for most of the low-mortality countries. Moreover, choosing the relational populations based on closest lifespan disparity made the choice of reference populations more scientific.
Presented in Session P1. Postercafe