See #1262. This has the same changes with the history rewritten a bit.
TST: Add tests for failing ARMA(0,0) case.
ENH: Add support for ARMA(0,0) models
I'm going to go ahead and merge this, so it doesn't sit any longer. cc @bashtage @josef-pkt
Nevermind. AIC and BIC don't match gretl. They use OLS for this case and return an "estimated using least squares ( = MLE)." Should we just return a regression results instance if the order is (0,0)?
I guess it could be nice to have the prediction dates magic...
To muddy the waters, Stata returns the same AIC and BIC we use ie., the ARMA definitions.
R's base arima uses the same AIC as we do (the ARMA one), and I suppose we want these to be comparable, so we should just stick with what we have in this PR I think.
I think the IC situation should probably be re-considered someday - but in a different PR after discussion.