RSS of restricted and unrestricted models

[simondenysketels]

Hello

I've estimated two models with the same sample, an ARMA(6,7) and an ARMA(7,7). My next step would be to perform an F-test to check the significance of the extra variable.
I always thought that by definition, a bigger model has a better fit and thus a bigger RSS. Now in this particular case, this isn't so, which of course renders my F-test completely useless.
What am I missing here? (I'm studying "econometrics: time analysis" right now, so I'm pretty new to this field)

Thank you very much
Simon

[startz]

Your instincts are correct.

ARMA models are nonlinear. This means that it is possible, although unlikely, that one of the estimates isn't quite complete. You might want to post your output and/or workfile together with the equations that are causing the problem.