I'm not sure what you mean by AR(p,q). Maybe ARMA(p,q)? In that case it
would normally be full-sample, neither recursive nor rolling. (Although
it wouldn't be a regression anyway, strictly speaking.)
As to automatic model selection. you would basically have to script the
loop yourself. There are too many possible ways how to choose a model
(which criterion to optimize) for gretl to guess which method you
prefer. (At least I guess that's the reason.)
Am 25.08.2010 22:36, schrieb Adrian Fath:
I'm running an AR(p,q) estimation and have two questions:
1) is the regression recursive or rolling (I cannot say by just looking
at the output)???
2) is there a way to let AIC decide how many lags to use???
Gretl-users mailing list