Forecasting Accuracy of a Multi-Country Macroeconometric Model for the Former Yugoslavia

Abstract

In this paper, the forecasting performance of a structural multi-country model for Slovenia, Serbia, Croatia and Bosnia and Herzegovina is compared to the forecasting ability of ARMA and VAR models. It turns out that in many cases the time series approaches deliver more accurate predictions of economic growth, inflation and the unemployment rate than the structural model which is based both on economic theory and on time series for the included countries. This shows that it is not so easy to beat time series models in forecasting. However, the usefulness of structural models should not be evaluated on the basis of their forecasting performance alone, since these models can in addition be used to investigate the macroeconomic effects of policy measures or of exogenous shocks.