GDP revisions are not cool: the impact of statistical agencies’ trade-off

Abstract

Official estimates of economic growth are regularly revised and therefore forecasts for GDP growth are done on the basis of ever-changing data. The economic literature has intensively studied the properties of those revisions and their implications for forecasting models. However, much less is known about the reasons for Statistical Agencies (SAs) to revise their estimates beyond the timeliness of their data collection. We show that SAs behave as risk managers who have an implicit interest (loss function) in not revising their initial GDP estimates too much, while they are much more open to revise GDP expenditure components over time. More than a curiosity, we exploit this resulting cross-correlation of revisions among GDP components to build a model to better forecast GDP.