Point and Density Forecasts for the Euro Area Using Bayesian VARs

Recent articles suggest that a Bayesian vector autoregression (BVAR) with shrinkage is a good forecast device even when the number of variables is large. In this paper we evaluate different variants of the BVAR with respect to their forecast accuracy for euro area real GDP growth and HICP inflation. We consider BVAR averaging, Bayesian factor augmented VARs (BFAVARs), and large BVARs, which differ in the way information is condensed and shrinkage is implemented. We find that: (a) large BVARs produce accurate point forecasts but show a poor performance when the entire density is considered; (b) BVAR averaging shows the opposite pattern; (c) BFAVARs perform well under both evaluation criteria; (d) choosing the degree of shrinkage optimally does not improve forecast accuracy; (e) all variants except the large BVAR tend to be well calibrated for inflation but poorly calibrated for real GDP growth; (f) these findings are robust to several features of the forecast experiment.