Zusammenfassung:
This paper derives forecasts for euro area real GDP growth based on a bottom up approach from the production side. That is, GDP is forecast via the forecasts of value added across the different branches of activity, which is quite new in the literature. Linear regression models in the form of bridge equations are applied. In these models earlier available monthly indicators are used to bridge the gap of missing GDP data. The process of selecting the best performing equations is accomplished as a pseudo real time forecasting exercise, i.e. due account is taken of the pattern of available monthly variables over the forecast cycle. Moreover, by applying a very systematic procedure the best performing equations are selected from a pool of thousands of test bridge equations. Our modelling approach, finally, includes a further novelty which should be of particular interest to practitioners. In practice, forecasts for a particular quarter of GDP generally spread over a prolonged period of several months. We explore whether over this forecast cycle, where GDP is repeatedly forecast, the same set of equations or different ones should be used. Changing the set of bridge equations over the forecast cycle could be superior to keeping the same set of equations, as the relative merit of the included monthly indictors may shift over time owing to differences in their data characteristics. Overall, the models derived in this forecast exercise clearly outperform the benchmark models. The variables selected in the best equations for different situations over the forecast cycle vary substantially and the achieved results confirm the conjecture that allowing the variables in the bridge equations to differ over the forecast cycle can lead to substantial improvements in the forecast accuracy.
