A multilevel functional data method for forecasting population, with an application to the United Kingdom

Han Lin Shang*, Peter Smith, Jakub Bijak, Arkadiusz Wiśniowski

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

20 Citations (Scopus)

Abstract

Cohort component models are often used to model the evolution of an age-specific population, and are particularly useful for highlighting the demographic component that contributes the most to the population change. Recently, most of the attention has been devoted to the estimation of four specific demographic components, namely mortality, fertility, emigration and immigration. Many methods take a deterministic viewpoint, which can be quite restrictive in practice. The statistical method that we propose is a multilevel functional data method, where both mortality and migration are modelled and forecast jointly for females and males. The forecast uncertainty associated with each component is incorporated through parametric bootstrapping. Using the historical data for the United Kingdom from 1975 to 2009, we found that the proposed method shows a good in-sample forecast accuracy for the holdout data between 2001 and 2009. Moreover, we produce out-of-sample population forecasts from 2010 to 2030, and compare our forecasts with those produced by the Office for National Statistics.
Original languageEnglish
Pages (from-to)629-649
Number of pages21
JournalInternational Journal of Forecasting
Volume32
Issue number3
DOIs
Publication statusPublished - Jul 2016
Externally publishedYes

Keywords

  • Age-specific population forecasting
  • Population projection matrix
  • Leslie matrix
  • Linear growth models
  • Functional data analysis
  • Functional principal component analysis

Fingerprint

Dive into the research topics of 'A multilevel functional data method for forecasting population, with an application to the United Kingdom'. Together they form a unique fingerprint.

Cite this