Updating function model: Model updating method transferable in a wider range of data sizes

Abstract

When data are available from two time points—older data with a larger number of observations and more recent data with a smaller number of observations—then model updating is utilised to take advantage of the different merits of each data set. However, the author's previous study demonstrated that conventional model updating methods—transfer scaling, joint context estimation, Bayesian updating, and combined transfer estimation—were inferior to models using only the more recent data. The present study examines an updating method that the author calls an ‘updating function model’ in which the parameters are assumed to follow the functions of gross domestic product per capita. The present study demonstrates that the updating function model often produces statistically significantly better forecasts than models using only the more recent data. The study extends the applicability of the model updating to cases in which the more recent time point has more observations than the older time point.