Review of Demographic Forecasting Methods

In the context of persistent global fertility decline and population aging, the need to improve demographic forecasting and modeling methods is increasing. According to the World Bank, the global total fertility rate declined from 5.5 in 1960 to 2.2 in 2023. In Russia the topic of demographic modeling and forecasting is particularly important due to the state objective of increasing the fertility rate. This paper provides a brief review of demographic forecasting methods — from classical econometric approaches to machine learning algorithms and agent-based modeling. The strengths and weaknesses of various model classes are assessed along with their limitations in the context of short-term and long-term forecasting. The paper sequentially examines several main groups of methods: classical time series models (Lee-Carter model, ARIMA specifications, ETS models), Bayesian probabilistic approaches, machine learning (ML) methods, reinforcement learning (RL) methods and agent-based modeling (ABM).Agent-based modeling occupies a special place in this review as it shifts the focus from extrapolation of aggregated trends to the reproduction of micro-level mechanisms. Another important part of the paper is reinforcement learning methods which allow purposeful formation of adaptive management policies in response to changes in the demographic, social and economic environment. This review is based on recent publications, including studies by Russian and foreign authors as well as classical works that laid the foundations of the methods under consideration.