Abstract
Revisions to percentage changes in the total index of industrial production (IP) contain systematic components and are correlated with other information available at the time the provisional estimates are prepared, indicating that the preliminary estimates are inefficient. In particular, revisions to percentage changes in IP (i) tend to be upward, (ii) contain substantial seasonality, (iii) generally increasse during periods in which IP is accelerating, and (iv) revisions to the same month tend to be in the same direction. Filters for the provisional estimates that exploit these inefficiencies are estimated, and the estimates generated from these filters pass the efficiencies are estimated, and the estimates generated from these filters pass the efficiency test.