PROBLEMS AND SOLUTIONS IN FORECASTING GEOGRAPHICAL POPULATIONS

This paper asks the question: how does the multistate population model need to be adjusted to provide forecasts of geographical populations? Following an exposition of the standard model, possible solutions to the problems posed by excessive number of parameters are discussed. Decomposition, aggregation and parameterization are described, drawing on some new results. Issues in the temporal forecasting of model components are outlined and the alternative approach of using a spatial interaction model is considered. The paper concludes by arguing that the design of forecasting models is a powerful learning device for both designers and users.