inverted MA roots close to 1
Posted: Mon Jun 01, 2015 11:14 pm
Dear all, are inverted MA roots close to one a problem for model validity or interpretation? If yes, what would be a recommended step (e.g. reducing MA terms etc). Some published results seem to indicate "no problem". EViews 7 User’s Guide II from Prof. Schwert on p. 100 explicitly states "...that the inverted roots have moduli very close to one, which is typical for many macro time series models". Contrary, other softwares such as R (auto-arima) would ignore these models due to "forecasting issues". What is right? Are "MA roots within the circle" sufficient? Thanks.