Regional Probabilistic Fertility Forecasting by Modeling Between-Country Correlations

Area of Science:

• Demography
• Statistical Modeling
• Population Studies

Background:

• The UN Population Division uses a Bayesian hierarchical model for probabilistic total fertility rate (TFR) projections.
• Current models project TFR for individual countries but lack joint projections for aggregates.
• There's a need for accurate probabilistic projections of aggregate TFRs for regions and trading blocs.

Purpose of the Study:

• To extend the existing Bayesian hierarchical model for joint probabilistic projection of aggregate TFRs.
• To develop a method for projecting TFR for any combination of countries.

Main Methods:

• Modeled correlation between country forecast errors using time-invariant covariates (contiguity, common colonizer, UN region).
• Incorporated the correlation model into the Bayesian hierarchical model's error distribution.
• Produced predictive distributions of TFR for UN regions (1990-2010).

Main Results:

• The enhanced model produced prediction intervals closer to nominal levels compared to the current model.
• Geographic proximity was identified as a significant factor in TFR forecast error correlation.
• The proportion of observed values within prediction intervals improved.

Conclusions:

• Geographic proximity substantially explains correlation in TFR forecast errors between countries.
• Accounting for this correlation significantly improves probabilistic TFR projections for aggregates.
• The enhanced model offers more reliable TFR projections for regions and other country groupings.