Forecasting using Bayesian VARs : a benchmark for STREAM

This study develops a suite of Bayesian Vector Autoregression (BVAR) models for the Maltese \neconomy to benchmark the forecasting performance of STREAM, the traditional macro-econometric \nmodel used by the Central Bank of Malta for its regular forecasting exercises. Three different BVARs \nare proposed, containing an endogenous and exogenous block, and differ only in terms of the cross- \nsectional size of the former. The small BVAR contains only three endogenous variables, the medium \nBVAR includes 17 variables, while the large BVAR includes 32 endogenous variables. The exogenous \nblock remains consistent across the three models. By using a similar information set, the Bayesian \nVARs developed in this study are utilised to benchmark the forecast performance of STREAM. In \ngeneral, for real GDP, the GDP deflator, and the unemployment rate, BVAR median projections for \nthe period 2014-2016 improve the forecast performance at the one, two, and four-step ahead horizons \nwhen compared to STREAM. However, the latter does rather well at annual projections, but it is \nbroadly outperformed by the medium and large BVARs.