This paper compares the forecasting performance of three different econometric models for the Eurozone and the USA: a vector
auto regression (VAR), a Bayesian vector auto regression (BVAR), and a structural vector error correction model (SVEC). The
forecast evaluation is based on 19 vintages of real time data for output, inflation rates, interest rates, the exchange rate
and the money stock from the fourth quarter of 2004 until the first quarter of 2010. The oil price is used as the only exogenous
variable in the model. Imposing a stringent set of long-run assumptions on the econometric model results in less accurate
forecasts. The difference is significant for several variables and forecast horizons. Reducing the comparison to data from
the pre-financial crisis period reduces the size of forecast errors but does not change the overall picture.
Forschungsbereich:Makroökonomie und öffentliche Finanzen