NONCOINTEGRATION AND CAUSALITY - IMPLICATIONS FOR VAR MODELING

Tests for cointegration and causality are used to investigate the interrelatedness of state, regional and U.S. non-agricultural employment. Tests for cointegration show that states and regions are generally not in 'equilibrium' with other states and regions or the U.S. economy as a whole. Using final prediction error (FPE) criterion, however, U.S. employment is shown to 'cause' state and regional employment in most cases. Thus, changes in U.S. economic activity appear to influence state and regional activity over time, but without maintaining equilibrium relationships. Experimental autoregressive forecasts of all 50 states, the District of Columbia and five regions suggest these results may have important implications in the specification of regional VAR models, particularly with respect to the inclusion and exclusion of variables under stationary and level approaches.