Measuring UK GDP Growth Data Uncertainty by Ana Galvao with James Mitchel

Economic statistics are prone to data uncertainty since they are subject to both sampling and non-sampling errors. GDP estimates, our focus, are regularly revised as new information is received and methodological improvements are made. We first show that these revisions matter, emphasising that they are time-varying and contain `news’. Although the Office for National Statistics, in the UK, emphasise that initial estimates of GDP values will be revised, it is the Bank of England that, uniquely in an international context, provide direct quantitative estimates of the likely ex ante uncertainty around these past GDP values. In this paper we compare these ex ante uncertainty estimates with ex post measures both to provide the first evaluation of the calibration of the Bank of England’s predictive densities for revised GDP growth values and to construct a new measure of data uncertainty, that captures only the unforecastable component to data revisions. To achieve this we propose a generic, loss-function based approach to measuring uncertainty; and show how density forecast calibration tests can then be constructed from this. We find that Bank of England’s point predictions better anticipate mature ONS growth estimates than the ONS’s own first releases; and that their density estimates are, on average, well-calibrated. But this masks temporal changes in predictive density performance. The direct econometric estimates of data uncertainty that we provide show that, like popular forward-looking (forecast-based) measures of macroeconomic uncertainty, data uncertainty also jumped at the onset of the 2008/2009 recession. This suggests that to better understand macroeconomic uncertainty it is important to monitor and track historical data uncertainty separately using estimates like the ones we propose.