Projection and Estimation of International Migration

Abstract

I propose techniques for improving both estimation and projection of international migration. By applying a Bayesian hierarchical modeling approach to net migration data, I produce projections of international migration that are global in scope and have well quantified uncertainty. My projections are of an appropriate form to be included as the migration component in probabilistic population projections, as has by done by Azose et al. (2016). (The current practice of the United Nations Population Division is to produce probabilistic population projections which include deterministic projections of migration.) The net migration model may be improved by incorporating a correlation matrix, but estimating such a matrix is difficult because the dimension of the matrix is large while the number of available data points is small. I demonstrate a method for estimating a correlation matrix which includes a prior belief that correlations which are large in magnitude are more likely among countries which are "close", either because of geographical or historical ties. Including correlations improves projections when net migration is aggregated across regions. I also propose a method for improving existing estimates of bilateral migration flows based on migrant stock data. A current state-of-the-art estimation method (Abel, 2013) relies on an unrealistic assumption that the total number of migrants is as small as possible, resulting in estimates with many structural zeroes. By weakening that assumption, I produce estimates of migration flows between all pairs of countries that allow for substantial return migration flows.