A BVAR MODEL FOR THE CONNECTICUT ECONOMY

A Bayesian vector autoregressive (BVAR) model is developed for the Con necticut economy to forecast the unemployment rate, nonagricultural em ployment, real personal income, and housing permits authorized. The mo del includes both national and state variables. The Bayesian prior is selected an the basis of the accuracy of the out-of-sample forecasts. We find that a loose prior generally produces more accurate forecasts. The out-of-sample accuracy of the BVAR forecasts is also compared wit h that of forecasts from an unrestricted VAR model and of benchmark fo recasts generated from univariate ARIMA models. The BVAR model general ly produces the most accurate short- and long-term out-of-sample forec asts for 1988 through 1992. It also correctly predicts the direction o f change.