BVAR Models with Economic Priors: An Application to the Propagation of U.S. Regional Cycles to Canada.

To overcome the over-parameterization problems typically associated with the estimation of large VAR systems, Litterman (1979, 1986) and Doan, Litterman, and Sims (1984) have proposed the inclusion of statistical a priori information. In this paper, we investigate how economic a priori information based on regional input-output tables and trade flows statistics could help estimate a large U.S.-Canadian regional model. Instead of relying on the usual Choleski factorization, we present the variance decomposition based on a national-regional unobservable variables model. Using monthly series (total employment, 1966:1-1986:12) on five Canadian regions and four U.S. ones, we are able to characterize the north-south propagation mechanism.