2026/05/26 by Charles Siriban, Aude Bernard, Jacques Poot +3 · 1 voice
Social Sciences · Economics, Econometrics and Finance · #Human Mobility and Location-Based Analysis #Urban, Neighborhood, and Segregation Studies #Regional Economics and Spatial Analysis
paper · doi:10.1080/00324728.2026.2645544
openalex publication_date 2026/05/26 · openalex created_date 2026/05/27 · openalex updated_date 2026/07/02
Recognizing that internal migration is an important source of error in population projections, we evaluate 21 forecasts from five classes of methods: (1) average of the most recent flows; (2) time-series econometric models with and without external factors; (3) machine-learning-based gradient boosting models; (4) multiplicative component and gravity-type models; and (5) average of all forecasts. We forecast bilateral interstate migration flows in Australia for the five years to mid-2016 and to mid-2023. No single model consistently outperforms the others in terms of bias, accuracy, and empirical coverage. Simple models perform as well or better than complex models. ARIMA models and their Bayesian equivalent perform well in both periods. Including control variables is no panacea, because their relationship with migration is flow and period specific. Similarly, globally trained machine learning models are not clear-cut alternatives, particularly when flow sizes vary widely. In contrast, the multiplicative component model suits contexts where the spatial structure of migration is stable, suggesting that internal migration forecasting requires context-specific approaches.