We evaluate the forecasting performance of six different models for short-term forecasting of Macedonian GDP: 1) ARIMA model; 2) AR model estimated by the Kalman filter; 3) model that explains Macedonian GDP as a function of the foreign demand; 4) small structural model that links GDP components to a small set of explanatory variables; 5) static factor model that links GDP to the current values of several principal components obtained from a set of high-frequency indicators; 6) FAVAR model that explains GDP through its own lags and lags of the principal components. The comparison is done on the grounds of the Root Mean Squared Error and the Mean Absolute Error of the one-quarter-ahead forecasts. Results indicate that the static factor model outperforms the other models, providing evidence that information from large dataset can indeed improve the forecasts and suggesting that future efforts should be directed towards developing a state-of-the-art dynamic factor model. The simple model that links domestic GDP to foreign demand comes second, showing that simplicity must not be dismissed. The small structural model that explains every GDP component as a function of economic determinants comes third, "reviving" the interest in these old-school models, at least for the case of Macedonia.
forecasting structural model principal component FAVAR static factor model Macedonia