VARMA VERSUS STEADY-STATE BVAR FOR MACROECONOMIC FORECASTING

This study examines whether Vector Autoregressive Moving Average (VARMA) models, which are theoretically attractive, can forecast macroeconomic data more accurately than the widely used Steady-State Bayesian Vector Autoregressive (BVAR) model. Previous research shows that VARMAs can outperform (frequentist) Vector Autoregressive (VAR) models in forecasting macroeconomic data. There is some evidence suggesting that this does not necessarily hold if the VARs are Bayesian. To investigate this further, an out-of-sample forecasting competition is conducted for a Swedish macroeconomic data set, comparing the point forecasts, interval forecasts, and density forecasts of VARMA, Steady-State BVAR, and a benchmark univariate Autoregressive (AR) model. The results show that, overall, in terms of point, interval, and density forecast performance, the Steady-State BVAR is superior to the VARMA and the benchmark. The VARMA does show improvement over the benchmark model; however, there are cases where the benchmark performs better. The results make a strong case for the continued use of the Steady-State BVAR, instead of VARMAs, for macroeconomic forecasting.