Forecasting migration in Europe: a comparison of traditional methods and the Bayesian approach

Project Manager: Jakub Bijak

The project is financed by the Foundation for Population, Migration and Environment (BMU/PME).

The research project is aimed at the assessment of existing methods of forecasting and development of an alternative forecasting tool to be applied in migration studies. International migration flows between selected European countries are investigated, the countries under study being the most important sources or destinations of migration in Europe.

The project consists of three stages:

1. Methodological assessment of the existing and potential methods of forecasting of international migration;
2. A thorough investigation of the applicability of Bayesian statistics to migration forecasting;
3. Comparison of the empirical results of the forecasts of international migration prepared with the innovative methods with the methods based on classical (frequentist) statistical paradigm and commonly used in forecasting up to date.

In the Bayesian approach certain a priori assumptions are made regarding the probabilities of changes in the independent variables. Bayesian methodology allows for incorporation of an expert opinion, or analogous information from the past experience of other countries, into statistical inference in a formal way through prior distributions of model parametres.

The proposed Bayesian forecasting model is going to be thoroughly evaluated and verified in several steps during the project, both by the means of ex ante and ex post errors, comparing them with the ones obtained from the traditionally used time series models. Robustness of the results against model assumptions and changes in the migration development scenarios are also going to be verified. It will be examined, whether the results significantly differ when other type of prior information on the parameters is applied or other type of interdependence between variables is assumed. Especially of interest is the ex post evaluation of robustness against discontinuity in trends: whether the models are 'able' to predict sudden changes of migration patterns. All these steps will allow for an evaluation of the applicability of the Bayesian methodology in forecasting international migration.