Forecasting industrial employment figures in Southern California: A Bayesian vector autoregressive model

Abstract

In this paper, we construct a Bayesian vector autoregressive model to forecast the industrial employment figures of the Southern California economy. The model includes both national and state variables. The root mean squared error (RMSE) and the Theil's U statistics are used in selecting the Bayesian prior. The out-of-sample forecasts derived from each model and prediction of the turning points show that the Bayesian VAR model outperforms the ARIMA and the unrestricted VAR models. At longer horizons the BVAR model appears to do relatively better than alternative models. A prior that becomes increasingly looser produces more accurate forecasts than a tighter prior in the BVAR estimations.