This is partly an R question and partly a stats question:

I am trying to do batch forecasts using the auto.arima function from the forecast package. I have over 1000 items to forecast so doing it by hand is not feasible.

The auto.arima function consistently gives me predictions whose deviation from the true values have a pattern-- the errors are always positive and many have a specific trend (for example, they look like a downward sloping line).

I read some things about how this is a "known issue" in R (#3 here http://www.stat.pitt.edu/stoffer/tsa2/Rissues.htm), but I'm still confused by how to fix it. The solution I've come across is to use the SARIMA function, but is there an auto.arima equivalent for this? If not, is there another way to do batch forecasts? Maybe there's a different way of dealing with prediction error autocorrelation?

  • 1
    $\begingroup$ Hello Olga, auto.arima actually estimates seasonal components too. For example: auto.arima(AirPassengers) returns a seasonal model... $\endgroup$ – DatamineR Jul 16 '13 at 17:48
  • $\begingroup$ Thanks, Johnny. I was looking at the third issue outlined here: stat.pitt.edu/stoffer/tsa2/Rissues.htm The author says that using sarima avoids these problems, but gives no explanation why or how. $\endgroup$ – Olga Mu Jul 16 '13 at 19:37
  • $\begingroup$ If the errors are always positive, something is seriously wrong. Are you able to create a reproducible example? $\endgroup$ – AdamO Dec 5 '17 at 22:39

Your Answer

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

Browse other questions tagged or ask your own question.