To me what is looking most exotic is the different orders of integration of your models, which you are assuming starting from 1 through 5. All asymptotic results regrading the distribution of the model parameters based on the fact that original DGP has exactly 1 as the order of integration, because most of the real life scenarios which are non-stationary in nature can be well approximated with that. Therefore perhaps usual t-values can not be justified once you cross the limit as 1.
Apart from that, in my belief you can just go ahead with different combinations of p and q parameters and choose the optimal one (based on some pre-fixed criteria like AIC/BIC or non-significance of model coefficients). However in each experiment you should fix the initial values of the time series and should keep it same for all experiment. Best regards, -- View this message in context: http://r.789695.n4.nabble.com/ARIMA-models-tp2957129p2964483.html Sent from the R help mailing list archive at Nabble.com. ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
