Subnational Population Projections by Age: An Evaluation of Combined Forecast Techniques

Abstract

Population projections commonly suffer from a decreasing accuracy with a higher degree of disaggregation. We suggest combining the cohort-component model and an averaged projection technique to improve precision of forecasts for small areas. Calculating ex-post population projections, we evaluate the precision and bias of the proposed method by contrasting error patterns of commonly used techniques using official population data from 46 districts of a German state for the years 1980–2013. In comparison to the individual methods considered, the combined approach results in the highest accuracy and lowest bias both for the total population and for age groups. The proposed method is also more robust regarding past growth.