Bayesian VARs: Specification Choices and Forecast Accuracy

In this paper we examine how the forecasting performance of Bayesian VARs is affected by a number of specification choices. In the baseline case, we use a Normal-Inverted Wishart prior that, when combined with a (pseudo-) iterated approach, makes the analytical computation of multi-step forecasts feasible and simple, in particular when using standard and fixed values for the tightness and the lag length. We then assess the role of the optimal choice of the tightness, of the lag length and of both; compare alternative approaches to multi-step forecasting (direct, iterated, and pseudo-iterated); discuss the treatment of the error variance and of cross-variable shrinkage; and address a set of additional issues, including the size of the VAR, modeling in levels or growth rates, and the extent of forecast bias induced by shrinkage. We obtain a large set of empirical results, but we can summarize them by saying that we find very small losses (and sometimes even gains) from the adoption of specification choices that make BVAR modeling quick and easy. This finding could therefore further enhance the diffusion of the BVAR as an econometric tool for a vast range of applications.