A New Dimension to Mortality Forecasts?

Benjamin Seligman, Stanford University
Shripad Tuljapurkar, Stanford University

Since 2000, most government actuaries have used the Lee-Carter model for forecasting mortality rates and life expectancy, replacing mortality scenarios with meaningful prediction intervals. These forecasts rely on a dominant, linear trend of declining mortality over several decades, an assumption that has been appropriate for the past century. We examined mortality change in recent decades using the Lee-Carter approach; we used forecast models fitted to data from 1950-1989 and compared their forecasts to actual change since 1990. The dominant long-term trend continues to accurately predict aggregate mortality and mortality among women. For males, this yields a forecast that significantly underestimates actual changes. While male mortality decline followed a linear trend, the rate of change and its age-pattern are different from that over the long-term. We discuss the reasons for these shifts in the context of the Lee-Carter model, and more generally in terms of drivers of the sex-differences in mortality.