Do macroeconomic variables help predict international migration? insights from bayesian VAR ‘general-to-specific’ modelling

Jakub Bijak, University of Southampton

The paper addresses the issue of forecasting international migration with the help of two important theory-based macroeconomic predictors: income differentials (a pull factor) and unemployment (a push factor). The analysis is performed within the Bayesian vector autoregression (BVAR) framework and follows the ‘from general to specific’ approach to model selection. In BVAR processes, predictive uncertainty stems not only from forecasting migration and its determinants, but also from the randomness of interactions between particular variables, embodied in the parameters of the forecasting model. The Bayesian paradigm also enables small-sample inference, while allowing to include prior information about the model structure. The discussion is illustrated by forecasts of long-term migration flows between Germany and Poland, Italy, and Switzerland, prepared for 2005–2015. Their results indicate a trade-off between the lack of a rationale for including macroeconomic determinants in the forecasts, and extremely large predictive uncertainty yielded by general BVAR models with additional variables.