Due to the phenomenon of revision, published macroeconomic data can never be regarded as final, because they are subject to continuous change, although the size of the change decreases over time. Our methodology is able to give an estimate of expected future routine revisions, based on the time series properties of the revisions and the data. We intend to filter out the revisions stemming from methodological changes. In addition – in line with the practice used in the literature – we highlight the effect of turning points. Taking these into account, we can observe a more significant systematic bias. We applied this methodology to the time series of Hungarian GDP. Based on our results, GDP growth in 2016 is expected to be subsequently revised upwards in future publications by approximately 0.2 percentage points. Considerable uncertainty surrounds our estimate: the 90 per cent confidence interval surrounding the expected revision is approximately ±0.5 percentage points. Our results are also confirmed by two robustness analyses. In the first analysis, instead of using indicators in the model to capture the effect of the turning points, we define turning points based on the output gap. For the second estimate, we filtered out the effect of the latest major methodology change, i.e., the introduction of ESA 2010 from the revisions. In both cases we obtained results that barely differ from the baseline estimate for the expected revision.
JEL codes: C22, C53
Keywords: revision, GDP, state-space model, Kalman filter