Forecast accuracy hardly improves with method complexity when completing cohort fertility

Abstract

Forecasts of completed fertility predict how many children will be born on average by women over their entire reproductive lifetime. These forecasts are important in informing public policy and influencing additional research in the social sciences. However, nothing is known about how to choose a forecasting method from a large basket of variants. We identified 20 major methods, with 162 variants altogether. The approaches range from naive freezing of current age-specific fertility rates to methods that use statistically sophisticated techniques or are grounded in demographic theory. We assess each method by evaluating the overall accuracy and if provided, uncertainty estimates using fertility data of all available birth cohorts and countries of the Human Fertility Database, which covers 1,096 birth cohorts from 29 countries. Across multiple measures of forecast accuracy, we find only four methods that consistently outperform the naive freeze rates method, and only two methods produce uncertainty estimates that are not severely downward biased. Among the top four, there are two simple extrapolation methods and two Bayesian methods. The latter are demanding in terms of input data, statistical techniques, and computational power but do not consistently complete cohort fertility more accurately at all truncation ages than simple extrapolation. This broad picture is unchanged if we base the validation on 201 United Nations countries and six world regions, including Africa, Asia, Europe, Latin America and the Caribbean, northern America, and Oceania.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 57 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.