Insee's projections for France form a tool that has taken on a central importance in the public debate in the country.
What is the benefit of probabilistic projections?The high and low scenarios proposed by Insee for France allow us to set the limits of the uncertainty, while the probabilistic projections incorporate the risk in different ways: there are no variants but rather a set of scenarios built on the basis of probability densities. The main advantage of these projections is being able to offer not only a central estimate but also a confidence interval for any derived indicator (for example, the proportion of women among the over 65s in 2070). Vianney Costemalle engages in this exercise for France. In addition to proving the feasibility of these projections by actually carrying them out, he shows some differences compared to the usual Insee projections. The central value of projected fertility for 2070 is the same (1.95 children per woman), but the uncertainty is higher: the 95% confidence interval, assimilated here to the gap between the high and low hypotheses, is [1.63; 2.26] compared with [1.8; 2.1] for the high and low scenarios. Conversely, the mortality scenarios are both more pessimistic and narrower: 88.4 years and 92.0 years for life expectancy at birth for men and women in 2070, plus or minus a year, compared with 90 and 93 years, plus or minus three years, in the high and low scenarios.How do we evaluate the projections? One way of evaluating past projections consists in comparing them with actual developments. Nico Keilman has shown in previous research that, for 40 years, the projections have not come close to reality, concluding that we need to make probabilistic projections (Keilman, 2008). Here, he proposes a method for evaluating this type of projection, and applies it to those of three countries, France, Norway and the Netherlands. This allows him to revisit the projections he participated in 10 years ago and to show that they turned out to be more accurate than official projections, except in the case of France where the adjustments made in 1999 and 2006 were not correctly taken into account in the estimation of the parameters. He also shows that the errors are more marked for certain age groups, either because there is more uncertainty here or because the adjustments related specifically to those ages.
How do we build the projections?The components method used in the projections consists in estimating, for each year, net migration by sex and age, deaths by sex and age on the basis of the mortality rates, and the total number of births on the basis of the number of women of childbearing age and the fertility rates by age.The method is very effective as the sex and age of the inhabitants are very easy to forecast: girls aged 10 in 2020 will become women aged 60 in 2070, if they are still alive. These very severe restrictions regarding sex and age enable us to develop population projections that are much more robust than other projections (for example, economic projections) and to propose...