Revision confidence limits for recent data on trend levels, trend growth rates and seasonally adjusted levels

It is generally realised that recent seasonally adjusted or trend values are liable to be revised, even without changes to the unadjusted data, as further data points are added. However, there are no generally accepted indications of the likely scale of such revisions. This paper describes a method of using an ARIMA model of the unadjusted series, which is normally produced as the first stage of seasonal adjustment with either X-12-ARIMA or Seats, as the basis of calculating confidence limits for the revisions. The method of calculation involves expressing the revision as a function of the future innovations of the ARIMA model. To a good approximation the relevant function is a linear combination; since the innovations are by definition independent, it is straightforward to calculate the appropriate revision variance. The adequacy of the linear approximation is confirmed in a sample case by a Monte Carlo simulation. The method is illustrated by application to actual series, including to UK data on unemployment and prices. It is shown that the results can be useful, among other things, in judging the reliability of recent apparent turning points and in assessing the value of forecast extension in seasonal adjustment. It also gives interesting results in the comparison of adjustments using Seats and X-12.