Calibrating Probabilistic Forecasts of Finnish Fertility on Past Forecast Errors

Ricarda Duerst, Max Planck Institute for Demographic Research
Jonas Schöley, Max-Planck Institute for Demographic Research
Julia Hellstrand, University of Helsinki
Mikko Myrskylä, Max Planck Institute for Demographic Research and London School of Economics and Political Science (LSE)

We present probabilistic forecasts of the Finnish Total Fertility Rate (TFR) from 2024 to 2070 calibrated on the historically observed distribution of forecasting errors. The forecasts come from two scenario-based models. The postponement time series model (PPS) assumes that fertility postponement will gradually decline and eventually stop. The second model is a naive freeze-rates approach to forecasting fertility. We propose a novel quantile-mapping approach whereby we map non-calibrated forecasted TFR paths from the PPS forecast model to a target distribution derived from historical out-of-sample forecasting errors. The validation shows that our TFR forecasts calibrated on historical data outperform the non-calibrated TFR forecasts in coverage and interval score metrics. The results demonstrate the efficacy of empirical error quantification and quantile-mapping in calibrating probabilistic demographic forecasts.

Keywords: Population projections, forecasts, and estimations, Fertility, Simulation , Data and Methods

See paper.