Evaluating Mortality Forecasts Using Taylor's Power Law

Christina Bohk, University of Rostock
Roland Rau, University of Rostock
Joel E. Cohen, Rockefeller University and Columbia University

In 1961, Taylor established a power law that describes a linear relationship of log variance to log mean for population density; Taylor's law has been verified for many species in ecology, and recently for Norway's population density in human demography. In this paper, we show that Taylor's law also appears to be a regular pattern in human mortality data and how it could be used to compare and evaluate mortality forecasts. To do this, we forecast mortality for twelve countries of the Human Mortality Database from 1991 to 2009, given data from 1965 to 1990, applying different approaches like the canonical Lee-Carter model, some of its extensions and our model. The results of these retrospective forecasts suggest that only recently developed approaches can capture dynamic changes of mortality appropriately and that they can, therefore, substantially reduce forecast errors in comparison to previous approaches.