Modelling and forecasting age-specific death rates at older ages

Modelling and forecasting age-specific death rates at older ages

The study was conducted by Marius-Dan Pascariu, currently employed by SCOR-Life, and coordinated by James Vaupel, Professor at the University of Southern Denmark.

Duration of the Project: 2016 – 2018


The project was aimed at understanding and modelling mortality evolution using mathematical/demographic models. For the world as a whole, life expectancy has more than doubled over the past two centuries. This transformation of the duration of life has greatly enhanced the quantity and quality of people’s lives. It has fueled an enormous increase in economic output and in population size, including an upsurge in the number of elderly. Understanding human mortality dynamics is of the utmost importance in the context of rapid aging and increasing length of life experienced by most populations nowadays. The research highlights new and innovative methods for estimating and projecting future mortality levels among humans.

Three studies were devised, which develop and analyze relevant statistical models for addressing uncertainty in future mortality. The first study develops a method for forecasting male and female life expectancy. To forecast female life expectancy, the method is based on analyzing the gap in life expectancy between women in a given country and women in record-holding countries. To forecast male life expectancy, the gap between male life expectancy and female life expectancy in a given country is analyzed. The second study explores a new approach inspired by indirect estimation techniques applied in demography, which can be used to estimate full life tables at any point in time, based on a given value of life expectancy at birth or at any other age. The third study makes use of the statistical properties of a probability density function to estimate the distribution of deaths in a population in the future. Time series methods are used to forecast a limited number of central statistical moments and then reconstruct the future distribution of deaths using the predicted moments. The density function is estimated using the maximum entropy approach. The results show that mortality modelling can be tackled from different perspectives and higher accuracy of the future trajectories can be obtained when compared with the more traditional extrapolative methods based on age-specific death rates or probabilities.