Small Scale Bayesian VAR Modeling of the Japanese Macro Economy Using the Posterior Information Criterion and Monte Carlo Experiments

We construct Bayesian vector autoregressive (BVAR) models optimized by the Posterior Information Criterion (PIC), in which hyper-parameters are data-determined in the same way as the lag length and trend order. We also assess the performance of the selected models by one-step ahead forecasts using historical data and Monte Carlo experiments. The results suggest that the selected models have a superior performance in forecasting as compared with ordinary VAR models.