Prior selection for panel vector autoregressions

Abstract

Bayesian shrinkage priors have been very popular in estimating vector autoregressions (VARs) of possibly large dimensions. Many of these priors are not appropriate for multi-country settings, as they cannot account for the type of restrictions typically met in panel vector autoregressions (PVARs). With this in mind, new parametric and semi-parametric priors for PVARs are proposed, which perform valuable shrinkage in large dimensions and also allow for soft clustering of variables or countries which are homogeneous. The implication of these new priors for modeling interdependencies and heterogeneities among different countries in a panel VAR setting, is discussed. Monte Carlo evidence and an empirical forecasting exercise show clear and important gains from the new priors compared to existing popular priors for VARs and PVARs.