In dem Papier, das ich zusammen mit Florian Huber von der Oesterreichischen Nationalbank geschrieben habe, geht es um die Prognose von Rezessionen mittels eines globalen vektorautoregressiven Modells (GVAR).
Das (kurze) ganze Papier kann hier runtergeladen werden.
Wir motivieren die Studie folgendermaßen:
"Forecasting economic turning points is essential for optimally designing stabilization policy because of the lagged economic impact that scal and monetary policy stimuli. Consequently, the issue has been addressed in numerous studies (see, e. g., Zellner et al., 1991; Canova and Ciccarelli, 2004). In a globalized world, recessions are often triggered by external events. An example is given by the nancial crisis of 2008/09 that pushed many economies into a recession but originated, as many belief, from the burst of the U.S. housing market. This leaves room for global models to outperform country-specific models in forecasting turning points."
... und schlussfolgern am Ende:
"Using a large international panel of macroeconomic data, we have shown that GVAR models produce more accurate forecasts of recession probabilities than country-specific BVARs or univariate models. This result holds for most individual countries and forecast horizons as well as for several country groups. Thus, in a globalized world it might pay off to take information from other economies into account to improve the ability to forecast the future business-cycle regime. We anticipate that performance gains are even higher in the case of event forecasts when the event definition involves variables from more than one country (e.g. a recession in six of the G7 countries).
Since the focus of this paper is to study the gains from moving from country-specific models to a global model, we concentrate on relatively tractable linear benchmark models. At the same time, we acknowledge that other classes of non-linear models such as Markov-switching models can be superior tools to forecast business-cycle turning points (e.g., Kim and Nelson, 1998). Thus, it would be an interesting extension of our research to see whether taking the international dimension into account pays off also when comparing country-specific Markov-switching models with a Markov-switching GVAR model (as, e. g., in Binder and Gross, 2013)."
Es gibt noch ein zweites Papier zur Prognosegüte von GVAR-Prognosen, das das Modell ausführlicher beschreibt. Auch das sollte in ein paar Tagen online sein.