Data-based priors for vector error correction models

Abstract

We propose two data-based priors for vector error correction models. Both priors lead to highly automatic approaches which require only minimal user input. An empirical investigation reveals that Bayesian vector error correction (BVEC) models equipped with our proposed priors turn out to scale well to higher dimensions and to forecast well. In addition, we find that exploiting information in the level variables has the potential for improving long-term forecasts. Thus, working with VARs in first differences may ignore valuable information. A simulation study reveals that it is beneficial, in terms of estimation accuracy, to use BVEC in the presence of cointegration. But if there is no cointegration, the proposed priors provide a sufficient amount of shrinkage so that the BVEC model has a similar estimation accuracy compared to the Bayesian vector autoregressive (BVAR) estimated in first differences.