Title: Google data in bridge equation models for GDP
Authors: Thomas Goetz - Deutsche Bundesbank (Germany) [presenting]
Thomas Knetsch - Deutsche Bundesbank (Germany)
Abstract: There has been increased interest in the use of Big Data in general, and of Google Search data in particular, when it comes to forecasting macroeconomic time series such as private consumption, unemployment or inflation. However, applications on forecasting aggregate GDP, one of the variables predominantly focused on in central Banks, are rather rare. We incorporate Google Search data into a set of Bridge Equations, one of the workhorse models for short-term predictions of German GDP in the Deutsche Bundesbank. We show precisely how to integrate these big data information, emphasizing the appeal of the underlying model for this application. Naturally, the choice of which Google Search terms to add to which equation is crucial not only for the forecasting performance itself, but also for the economic consistency of the implied relationships. We use different ways of selecting the Google Search terms: subjectively, Google Correlate, based on a recursive out-of-sample forecast exercise, by extracting common unobserved factors, and a LASSO-approach. In a pseudo-real time forecast analysis we compare these alternatives in terms of their forecast accuracy for aggregate GDP and various disaggregate time series. We find that there are indeed sizeable gains possible from using Google Search data, which are likely to increase only further in the future, i.e. with ever more data.