An eigenvector spatial filtering contribution to short range regional population forecasting

A method to improve regional forecasts by incorporating spatial autocorrelation in a generalized linear mixed model framework coupled with eigenvector spatial filtering is proposed.

Abstract

Statistical space-time forecasting requires sufficiently large time series data to ensure high quality predictions. The dominance of temporal dependence in empirical space-time data emphasizes the importance of a lengthy time sequence. However, regional space-time data often have a relative small temporal sample size, increasing chances that regional forecasts might result in unreliable predictions. This paper proposes a method to improve regional forecasts by incorporating spatial autocorrelation in a generalized linear mixed model framework coupled with eigenvector spatial filtering. This methodology is illustrated with an application of regional population forecasts for South Korea.