Generalizing the Bayesian Vector Autoregression Approach for Regional Interindustry Employment Forecasting

The Bayesian vector autoregression (BVAR) employment-forecasting approach is generalized using data for the state of Georgia. This study advances previous regional BVAR approaches by (a) incorporating regional input-output coefficients instead of national coefficients, (b) using the coefficients both to specify the prior means in one model and to weight the variances of a Minnesota-type prior in a second model, and (c) including final-demand effects and links to national and world economies. Out-of-sample forecasts produced by the generalized BVAR models are compared to forecasts produced from an autoregressive model, an unconstrained VAR model, and a Minnesota BVAR model.