Dynamic forecast of irregular mortality developments within a Bayesian framework
Christina Bohk, University of Rostock
Roland Rau, University of Rostock
Forecasting irregular mortality developments is challenging, especially when longtime mortality trends change in the forecast years. Many countries in Central and Eastern Europe experienced an unsteady mortality development during the last 50 years, often including periods of stagnating and even decreasing life expectancy at birth; for instance, male life expectancy in Hungary decreased between the mid 1960s and the early 1990s, and strongly increased thereafter. As many mortality forecasting approaches extrapolate past trends, they fail to predict such trend changes. We try to overcome these problems with our novel mortality forecasting approach, which combines objective and subjective information in a Bayesian framework, i. e. we (1) use rates of mortality improvement (instead of death rates) to capture dynamic mortality developments, and we (2) can optionally complement a mortality trend in a country of interest with those of selected reference countries. These methodological refinements enable us to (1) incorporate flexible mortality dynamics and to (2) supplement and/or adjust them with expert judgment. In addition to a prospective forecast until 2050, we demonstrate in a retrospective application for Hungary that our model would have estimated Hungarian life expectancy more accurately than the original Lee-Carter model and two of its refinements proposed by Renshaw and Haberman: While the other applied models underestimate the progress in Hungarian life expectancy in 2009, after only 20 forecast years, by 4 to 5 years for women and by 6 to 8 years for men, we reduce this forecast error with our model to 2 years (for women and men) by only using the rates of mortality improvement; we then further improve the forecasting performance of our model by complementing the Hungarian mortality trend with that of West Germany, so that our forecasts exceed Hungarian life expectancy in 2009 by only 1 year for both sexes.