LATE FORECASTS AND EARLY REVISIONS OF UNITED-STATES GNP

We consider a sequence of four-monthly predictions and estimates of quarterly U.S. real gross national product. The late forecasts are released in the final month of a quarter, and an initial estimate and two subsequent revisions are published in the following three months. The question of efficiency is analyzed through considering the earlier figures as forecasts of this last number. The strongest indication of inefficiency appears in a nonlinear relationship between the third published number and each of the two previously reported estimates.