Drawing on both the Bayesian VAR and vector error corrections (VEC) literature, the author specifies the baseline model as a Bayesian VEC. The author documents the model's forecasting ability over various periods, examines its impulse responses, and considers several reasonable alternative specifications. Based on a root-mean-square-error criterion, the baseline model works best, and the author concludes that this model holds promise as a workhorse forecasting tool.

View the Full Working Paper