Application of the Kalman filter to revisions in monthly retail sales estimates

This paper applies the Kalman filter to improve upon published preliminary estimates of monthly retail sales, using ARIMA model projections as an alternative information source. Published revisions of retail sales estimates showed systematic patterns. Consequently, the filter problem is specified to incorporate models of the data revision processes as well as to treat contemporaneous covariances among residuals of the data revision models and the ARIMA retail sales models. The resulting specification, which is sufficiently general to be applicable to other bodies of data, yields considerable success in reducing the often substantial errors in the early retail sales estimates.