THE ACCURACY OF MACROECONOMIC FORECASTS BASED ON BAYESIAN VECTORIAL-AUTOREGRESSIVE MODELS. COMPARATIVE ANALYSIS ROMANIA-POLAND

The aim of this research is to make predictions for macroeconomic variables like inflation rate, unemployment rate and exchange rate for Romania and Poland using BVAR models. The one-step-ahead forecasts cover the horizon 2011-2013. Direct forecasts were developed using three types of priors for data covering the period from 1990 to 2012: Minnesota priors, non-informative priors and natural-conjugate priors. The forecasts’ accuracy assessment based on generalized forecast error of second moment put in evidence the superiority of Poland’s predictions based on a BVAR(2) model, compared to Romania’s ones based on a BVAR(4) model for differenced and stationary data series. For inflation rate the forecasts are rather inaccurate, but for Poland the Minnesota priors and for Romania the non-informative priors determined the most accurate predictions for unemployment rate and exchange rate on the horizon 2011-2012. It is very likely that these types of priors generate the best forecasts in 2013