EN
The paper compares three groups of methods used for best dynamic factor model selection for forecasting: modified information criteria, methods exclusively based on ex post forecasts analysis and mixed algorithms. It searches for the approach that delivers best out-of-sample forecasts according to mean square error measure. The analysis utilizes both Monte Carlo generated samples as well as real time series used for forecasting consumer inflation in Poland. Results show that best forecasts are obtained from the modified information criteria proposed by Groen and Kapetanios, whereas the methods that employ ex post forecasts from rolling windows usually give the worst predictions.