AStA Wirtschafts- und Sozialstatistisches Archiv

National Accounts are essential to macroeconomic analyses. In Germany, they are published quarterly by the statistical office (Statistisches Bundesamt). Since many source data necessary to compile them are still missing when the first dataset is released, the national accounts will be revised several times thereafter. Three and a half years after the end of each year this revision process ends, and the data are considered as final. These regular revisions interfere with benchmark revisions which take place approximately in a five years interval and are deemed to overhaul the entire system of national accounts. Observable is only the magnitude of total revisions, i.e. the sum of current revisions and benchmark revisions. However, to get hints at ways to improve the work of the statistical office and the legal framework for collecting data, one should look at the magnitude of current revisions. This paper proposes a simple approach to separate current revisions from benchmark revisions. In a next step, it analyses whether the current revisions of GDP, its main demand side components, employment, and labor productivity show some statistical regularities. The paper finds such regularities in terms of biasedness of the first release, autocorrelation of the revisions, and covariations of revisions with the size of the data. These findings may hint at possibilities to reduce the extent of revisions by making a better use of the information available. Finally, the paper shows that reducing the need of revisions would pay off in terms of a higher accuracy of the forecasts that are built on national accounts data.