Using Efficiency Tests to Reduce Revisions in Panel Data: The Case of Wage and Salary Estimates for U.S. States

This paper studies the preliminary estimates of state wage and salary growth produced by the U.S. Bureau of Economic Analysis (BEA), investigating whether they may be improved using basic efficiency tests. State wage and salary growth can be thought of as the sum of two components, an aggregate component equal to the currently prevailing national growth rate, and a state specific component, which we call the idiosyncratic variation in the estimates. BEA currently uses extrapolation techniques to compute the idiosyncratic component of the preliminary estimates, and this paper demonstrates some limitations to this approach. While revisions to BEA’s preliminary numbers are currently predictable, some simple regression-based generalizations of the extrapolation approach can reduce this predictability, cutting down mean squared and mean absolute revisions. While these reductions in revisions are not large, we can reject at conventional significance levels the hypothesis that they are equal to zero.