A New approach to Probabilistic Population Forecasting with an Application to Estonia

Abstract

The cohort component method is the approach most commonly used to generate population projections and Bayesian methods are now being applied to this approach in order to generate probabilistic population forecasts. This paper describes  an alternative to the Bayesian approach that is less complicated by showing how the Espenshade-Tayman method for transferring uncertainty can be applied to a projection based on the cohort component method. Using the Espenshade-Tayman method, we show how measures of uncertainty can be applied to existing population forecasts using Estonia  as a case study. The measures of forecast uncertainty are relatively easy to calculate and meet several important criteria used by demographers who routinely generate population forecasts. This paper applies the uncertainty measures to a population forecast based on the Cohort-Component Method, which links the probabilistic world forecast uncertainty to demographic theory, an important consideration in developing accurate forecasts. We applied this approach to world population projections and compared the results to the Bayesian-based probabilistic world forecast produced by the United Nations, which we found to be similar but with more uncertainty than found in the latter. We made a similar comparison in regard to sub-national probabilistic forecasts and found  our results to be similar with Bayesian-based uncertainty measures. These results suggest that the probability forecasts produced using our approach for Estonia are consistent with knowledge about forecast uncertainty. We conclude that this new method appears to be well-suited for developing probabilistic world, national, and sub-national population forecasts.

Keywords: ARIMA, Bayes, Cohort Component Method, Espenshade-Tayman Method, Fundamental Population Equation, Superpopulation