Revisions to the Czech National Accounts: Properties and Predictability

Frequent revisions to GDP and its components cause policymakers to face considerable uncertainty about the current state of the economy. In this paper, we provide stylized facts about the magnitude of revisions to the Czech national accounts. Using data over the 2003–2012 period, we find that the revisions are rather large. Revisions to real GDP growth are on average 1.4 for the annualized quarterly growth rate and 0.7 percentage points for the annual growth rate. Revisions to other variables are even larger: the average size of the revisions ranges from 1 to 12 percentage points for annualized quarterly growth rates and from 0.5 to 4 percentage points for annual growth rates. We investigate whether the revisions could have been predicted using the information available at the time of announcement. We find evidence for in-sample predictability for most of the variables, suggesting that the first releases of these variables are not efficient predictors of the actual values. In a real-time out-of-sample exercise, however, we find that the revisions to real GDP, gross fixed capital formation, and government consumption are not predictable. Only revisions to the GDP deflator can be predicted with substantial gains relative to zero-revisions forecasts.