Anticipating Continued Global Fertility Decline via Neural Forecasting

Daniel Ciganda, Facundo Morini, Francisco Piriz, Henrik-Alexander Schubert, Ugofilippo Basellini, Mikko Myrskylä · 2026

Read Anticipating Continued Global Fertility Decline via Neural Forecasting by Daniel Ciganda, Facundo Morini, Francisco Piriz, Henrik-Alexander Schubert, Ugofilippo Basellini, Mikko Myrskylä (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-23858.txt.

Introduction

Daniel Ciganda and colleagues forecast world fertility with NeuralTFR, a neural network that pools birth histories from 196 countries and territories. It learns shared demographic momentum across nations. Tested on 2009 through 2023, it beats both a naive drift baseline and the United Nations BayesTFR model on point forecasts. Its uncertainty calibration stays competitive. Looking to 2040, it sees low and very low fertility spreading wider than BayesTFR suggests. That implies weaker backing for a near term leveling off, though not the steepest decline paths in circulation.

Catalog record

CREATORDaniel Ciganda, Facundo Morini, Francisco Piriz, Henrik-Alexander Schubert, Ugofilippo Basellini, Mikko Myrskylä
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54479
LICENSEhttp://creativecommons.org/licenses/by-sa/4.0/
RIGHTSCC BY SA
SOURCEarXiv

Source: https://arxiv.org/abs/2605.23858

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Anticipating Continued Global Fertility Decline via Neural Forecasting Daniel Ciganda, Facundo Morini, Francisco Piriz, Henrik-Alexander Schubert, Ugofilippo Basellini, Mikko Myrskylä Abstract The accelerating shift toward low and ultra-low fertility has intensified the debate over whether countries now undergoing rapid decline are approaching stabilization or entering a more persistent low-fertility regime. Existing projection systems answer that question differently because they embed different assumptions about recovery and about the role of external drivers. To provide an empirical benchmark in this debate, we introduce NeuralTFR, an endogenous global forecasting framework based on a recurrent neural network. Drawing on a harmonized panel of historical fertility series from 196 countries and territories, the model pools cross-country information to learn demographic momentum and generate empirical prediction intervals via multi-quantile regression. Evaluated on a held-out period (2009–2023), NeuralTFR achieves lower point-forecast errors than a Naive Drift baseline and BayesTFR, the United Nations’ Bayesian Hierarchical Model, while maintaining competitive uncertainty calibration. In forward projections to 2040, NeuralTFR points to broader exposure to low and very low fertility than BayesTFR, suggesting weaker support for near-term stabilization while still falling short of the most severe decline paths predicted by the Global Burden of Disease project. Keywords— Total Fertility Rate, Global Neural Forecasting, Recurrent Neural Networks, Quantile Regression, Probabilistic Forecasting, Continued Fertility Decline 1 1 1 An interactive visualizer for the projection comparisons is available at https://famori.github.io/NeuralTFR/ . 1 Introduction The accelerating global transition toward low and ultra-low fertility poses profound challenges to economic growth, welfare systems, and geopolitical stability. Anticipating the depth and duration of this demographic shift depends upon robust projections of the Total Fertility Rate (TFR), a central challenge in demographic forecasting. A primary controversy in this context revolves around whether countries currently experiencing rapid declines will eventually stabilize and undergo a moderate recovery, or whether they are entering a more persistent low-fertility regime ( Basten et al. 2014 , see, e.g.,) . Global demographic forecasting is currently shaped by three major institutional approaches: the United Nations Population Division ( United Nations 2024 ) , the Wittgenstein Centre for Demography and Global Human Capital ( KC et al. 2024 ) , and the Global Burden of Disease (GBD) project ( GBD 2021 Fertility and Forecasting Collaborators 2024 ) . These paradigms differ markedly in both methodology and projected outcomes, reflecting different assumptions about post-transitional fertility, the role of expert judgment, and the importance of exogenous drivers. These disagreements are not simply methodological. They reflect a fundamental gap in demographic knowledge: the post-transitional fertility landscape has been observed in only a handful of countries over a limited period ( Goujon 2025 ) , so the assumptions that drive long-run projections operate with limited empirical guidance. This raises a natural question: what do recent fertility trajectories suggest when projections depend less on recovery assumptions, external covariates, and expert adjustment? To address that question, we introduce NeuralTFR, a global forecasting framework for the Total Fertility Rate built around an encoder-decoder architecture with Gated Recurrent Units (GRUs). Our approach formulates demographic projection as an autoregressive sequence-to-sequence task for forecasting multiple steps ahead. Rather than fitting isolated time-series models for each country, it uses a cross-series learning strategy. By jointly training a single model on a harmonized panel of historical fertility series from 196 countries and territories, NeuralTFR extracts shared, non-linear dynamics of the fertility transition directly from observed trajectories. The decoder is trained via multi-quantile regression, allowing the framework to generate empirical prediction intervals without relying on restrictive parametric assumptions or the computational burden of traditional Bayesian models. By isolating the endogenous momentum of historical fertility trends, this framework yields two primary contributions: First, methodologically, we show that a fully endogenous deep learning approach can perform strongly in demographic forecasting. Here, endogenous means that the model conditions on past fertility trajectories, without relying on exogenous socioeconomic covariates or expert adjustment, and that multi-step forecasts are generated autoregressively from the model’s own learned dynamics rather than from externally imposed recovery paths. Evaluated on a fully held-out period (2009–2023), NeuralTFR achieves lower point-forecast errors than a Naive Drift benchmark and the UN’s BayesTFR model, while maintaining competitive empirical interval coverage. Second, substantively, we provide a purely data-driven assessment of global fertility trajectories out to 2040. Our results suggest broader exposure to low and very low fertility than the United Nations projects, while still falling short of the most severe decline paths projected by the GBD framework. In that sense, when the global set of fertility trajectories is allowed to speak for itself, it offers limited support for either a near-term soft landing or the most severe decline scenarios currently published. 2 Previous Work Global demographic forecasting is currently dominated by three major institutional approaches. These models diverge significantly in their results, reflecting different ways of specifying post-transitional fertility and its long-run evolution. The first, and most widely adopted, is the probabilistic framework used by the United Nations Population Division ( United Nations 2024 ) , which relies on the Bayesian Hierarchical Model (BayesTFR) developed by Raftery, Ševčikova, Alkema and colleagues ( Alkema et al. 2011 ; Raftery et al. 2014 ; Raftery and Ševčíková 2023 ) . BayesTFR remains the central reference framework because it provides coherent probabilistic projections, encodes demographic transition structure, and has been extensively used in official projection practice. This model partitions the fertility transition into distinct phases. Its post-transition phase (Phase III) models fertility as an autoregressive AR(1) process that reverts to a country-specific long-term asymptote. Because this asymptote is informed by a global prior derived from countries that have previously experienced fertility recoveries, the model builds in eventual stabilization and a mean-reverting rebound. For instance, the UN explicitly assumes that fertility in countries with already-peaked populations will gradually rebound to an average of 1.4 births per woman by 2100 ( United Nations 2024 ) . Reflecting this expectation of recovering fertility, the UN anticipates the most conservative demographic decline, forecasting a global population of 10.2 billion by 2100. The second approach, developed by the Wittgenstein Centre for Demography and Global Human Capital (IIASA/VID), uses multi-state cohort-component models that emphasize educational stratification ( Lutz et al. 2014 ; Lutz et al. 2018 ; KC et al. 2024 ) . To define long-term fertility asymptotes, this model relies on panels of demographic experts rather than automated time-series extrapolation. Drawing on concepts like the low-fertility trap ( Lutz et al. 2006 ) , the IIASA model anticipates the persistence of low fertility and projects stabilization at levels generally lower than those assumed by the UN. Their most recent assessment (WIC2023) remains close to the United Nations over much of the overlapping horizon while projecting a global population of 9.9 billion by 2100 ( KC et al. 2024 ) . However, this approach depends heavily on subjective expert elicitation to set long-term bounds. A third approach was developed by the Global Burden of Disease (GBD) project at the Institute for Health Metrics and Evaluation (IHME). The GBD model projects fertility using an ensemble approach with four covariates: female educational attainment, contraceptive met need, under-5 mortality, and population density in habitable areas ( GBD 2021 Fertility and Forecasting Collaborators 2024 ) . Assuming these socio-economic and environmental drivers will continue to suppress birth rates,

Read the complete text