News or Noise? An Analysis of GNP Revisions

This paper studies the nature of the errors in preliminary GNP data, It first documents that these errors are large. For example, suppose the prelimimary estimate indicates that real GNP did not change over the recent quarter; then one can be only 80 percent confident that the final estimate (annual rate) will be in the range from -2.8 percent to +2.8 percent. The paper also documents that the revisions in GNP data are not forecastable, This finding implies that the preliminary estimates are the efficient given available information. Hence, the Bureau of Economic Analysis appears to follow efficient statistical procedures, in making its preliminary estimates.