The possibility to treat missing observations is mentioned explicitly in Giannone, Reichlin and Sala JME (2008) for the case of forecasting the factors when some of the time series from which they are extracted are missing in the context of nowcasting GDP using a DFM model. I would also be interested to know if this is also possible to be done backwards, that is to extract the factors in the part of the sample where some of the time series are missing, and how (I guess the way would be the same no matter where one goes, forward or backwards). Many thanks again. Andreas