I'm using the auto.arima function of theforecast package. I would like to perform the model selection using a custom metric instead of the usual information criteria. Is there a way to do that in auto.arima or I have to create my own solution in another way?

  • $\begingroup$ stats.stackexchange.com/questions/281646/… comments on some of the critical assumptions which limit the actal model selection. What metric are you considering and why ? $\endgroup$ – IrishStat Jun 30 '17 at 8:26
  • $\begingroup$ I have added experimental support for this in fable::ARIMA() via the selection_metric argument: github.com/tidyverts/fable/issues/280 It accepts a function which computes a metric from the stats::arima() output object, which is minimised to give the best model. $\endgroup$ – Mitchell O'Hara-Wild Jun 19 at 6:20

No, there is no way to do this in auto.arima(). You will need to replicate the code and change the information criteria to whatever metric you want to use.

(The same holds for the ARIMA() function in the successor fable package.)

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.