No. 520: Estimating Time-Variation in Measurement Error from Data Revisions: An Application to Forecasting in Dynamic Models
George Kapetanios ,
Queen Mary, University of London
October 1, 2004
Over time, economic statistics are refined. This means that newer data is typically less well measured than old data. Time variation in measurement error like this influences how forecasts should be made. We show how modelling the behaviour of the statistics agency generates both an estimate of this time variation and an estimate of the absolute amount of uncertainty in the data. We apply the method to UK aggregate expenditure data, and illustrate the gains in forecasting from exploiting our model estimates of measurement error.
J.E.L classification codes: C32, C53
Keywords:Forecasting, Data revisions