We study whether the accuracy of real-time estimates of the output gap produced by the OECD has improved over time by examining a panel dataset on real-time output gap revisions for 15 countries from 1991 Q1 - 2005 Q4. We use a simple panel data regression and a state space model, with common and country-specific factors, to assess whether the mean of the absolute value of output gap revisions has changed. We also apply a Mincer-Zarnowitz regression to determine whether the composition of measurement errors has shifted. Surprisingly, we do not find evidence that the size of output gap revisions has decreased over time; on the contrary, they seem to have increased. But the fraction of measurement errors due to omitted contemporaneous information has declind. This suggests that while the OECD's accuracy in output gap measurement has failed to improve, its estimates are becoming less noisy.
Output gap, real-time data, dynamic common factor model