Forecasting the Swiss economy with a small GVAR model

Abstract

We compare the forecast performance of a small global vector autoregressive (GVAR) model comprising Switzerland, the euro area, the US and Japan with two alternative models. The GVAR model produces more precise forecasts for Swiss inflation and real GDP growth from 2002 to 2011 than a cointegrating VAR model for Switzerland that is complemented with a simple VAR to predict the exogenous variables. Using predictions for the exogenous variables from the scenarios for the world economy that the SNB employs in its inflation forecasts, however, achieves lower forecast errors. This can be explained by the larger information set that enters these scenarios, in particular during the financial crisis. We conclude that the GVAR model is able to produce reasonable forecasts even if one is interested in a single economy that forms part of a global system.