Forecasting Migration: A Model Averaging Approach

Jesus Crespo Cuaresma, Wittgenstein Centre (IIASA, VID/ÖAW, WU)
Jakob Zellmann, Department of Statistical Sciences - University of Bologna
Juan Caballero, World Data Lab
Katharina Fenz, World Data Lab
Teodor Yankov, University of Oxford

Anticipating future migration dynamics is central to forming reasonable expectations of economic, demographic, and social developments. However, the discussion around which forecasting methods can provide the most accurate projections of global cross-country migration flows is still contested. As the complexity of migration processes has prevented the evolution of a unified theory of global migration, diverse statistical methods are applied to model migration. The literature typically distinguishes between causal and non-causal/autoregressive models. While the former links push and pull factors of migration to migration patterns, the latter extrapolates human mobility based on past trends. As there is high uncertainty about the adequacy of any individual approach to model migration, we propose a model averaging approach that incorporates the strengths of both causal and autoregressive models. In particular, we consider three non-causal models and a state-of-the-art gravity model to exploit the strengths of diverse modelling techniques. Using OECD data on bilateral migration flows, we conduct a pseudo out-of-sample validation exercise to gain insights into the predictive power of the individual models and their combinations. Our results suggest that model averaging does improve the predictability of migration models and thus provides evidence that model averaging is beneficial in empirical migration research.

Keywords: Bayesian methods , Data and Methods, Econometrics , International Migration

See extended abstract.