Comparison of Polish GDP forecasting quality with dynamic factor models and midas embedded with factor structure

The article is a continuation of the previous author’s papers (2007, 2009, 2012) devoted to the optimal methods of forecasting Polish macroeconomic variables, with the sample of GDP. The research was aimed at a comparison of the quality of nowcasts (”fore-casts” of the present time) and forecasts prepared with a dynamic factor model with mixed frequency and data gaps handling (MFDG-DFM) proposed by Mariano and Murasawa [2003] and MIDAS model augmented with factor structure (DFM-MIDAS), described for the first time in the paper of Marcellino and Schumacher [2008]. Mathematical backgrounds of both models were presented and a combination of Kalman filter and Maximum likelihood estimation was hinted as the estimation framework for both of them. The gained results show an advantage of Mariano and Murasawa approach in the field of the forecasts (approx-imately 15% more adequate forecasts for 2 and more quarters ahead) but this model is less adequate than the competitor for one quarter ahead forecasts and nowcasts