A review and evaluation of internal migration forecasting models

Area of Science:

• Demography
• Population Studies
• Spatial Analysis

Background:

• Internal migration significantly impacts population projection accuracy.
• Evaluating diverse forecasting methods is crucial for improving demographic modeling.
• Understanding migration patterns is key for effective policy and resource allocation.

Purpose of the Study:

• To assess the performance of 21 internal migration forecasting models across five categories.
• To identify the most accurate and reliable methods for bilateral interstate migration flow prediction in Australia.
• To determine if complex models offer advantages over simpler approaches in migration forecasting.

Main Methods:

• Comparative analysis of five classes of forecasting methods: flow averages, time-series econometric models, machine learning (gradient boosting), component/gravity models, and ensemble averages.
• Forecasting bilateral interstate migration flows in Australia for two distinct five-year periods (to mid-2016 and to mid-2023).
• Evaluation of models based on bias, accuracy, and empirical coverage.

Main Results:

• No single forecasting method demonstrated consistent superiority across all evaluation metrics.
• Simpler models, including ARIMA and its Bayesian variant, performed comparably or better than more complex methods.
• The effectiveness of control variables and machine learning models was context-dependent and not universally advantageous.
• The multiplicative component model showed utility in stable spatial migration contexts.

Conclusions:

• Internal migration forecasting necessitates context-specific strategies rather than a one-size-fits-all approach.
• The performance of forecasting models varies significantly based on data characteristics and temporal dynamics.
• Further research into context-aware model selection is recommended for improving population projection accuracy.