Demographic forecasting: approaches and developments
Heather Booth, Australian National University
Research interest in demographic forecasting is growing rapidly. The paper provides an overview of approaches and developments since 1980 using a simple organisational framework. Three approaches to forecasting demographic processes (principally mortality, fertility and migration) are discussed: statistical extrapolation, methods based on expectations (including expectations at the individual level and the informed opinions of experts at the population level), and theory-based structural modelling involving exogenous variables. The use of decomposition and disaggregation is also discussed. Cutting across these approaches is the number of factors (age, period and cohort) modelled: zero, one, two or three. Higher-factor models tend to be limited to extrapolation whereas the theoretical and expert-based approaches often use zero-factor models. Different approaches often produce substantially different forecasts. An important distinction between the four approaches is their ability to forecast change: in theory, extrapolative methods are least able to capture change while structural modelling is designed to do so. In practice, accuracy depends on the particular situation or trends, but it is not clear when a method will perform best. The estimation of uncertainty, central to probabilistic forecasting, is done by one of three methods: model-based ex ante error estimation, expert-opinion-based ex ante error estimation and ex post error estimation. These cover different sources of error and often produce different estimates. A major research advance is the evolution from traditional scenario-based deterministic population projections to probabilistic population forecasts based on stochastic population renewal. The approaches to population forecasting, demographic process forecasting and error estimation are closely linked; combinations of approaches are increasingly employed. The paper attempts to summarise developments and assess progress.
|
