I have an ARIMA(4,0,2) model that works well for 168-differenced data, that is, I fitted it to $Y_t-Y_{t-168}$.

Based on this, would it be a good idea to try to fit a Seasonal ARIMA(4,0,2)(0,1,0)[168] or a SARIMA(4,0,2)(0,1,1)[168]?

What would be a good way to decide the order of the polynomial in $B^{168}$ in front of the white noise terms?

  • $\begingroup$ I'm curious what type of business or natural time series has a 168 periodicity? $\endgroup$
    – Skander H.
    May 11, 2019 at 3:16
  • 1
    $\begingroup$ @Skander H. It’s hourly data of energy consumption with daily and weekly periodicity $\endgroup$ May 11, 2019 at 7:24

1 Answer 1


The first is the obvious candidate, since it's implied by the differencing (i.e. if you difference that first SARIMA you get the initial ARIMA model - with a zero mean).

However, that doesn't mean that the model with more parameters won't be noticeably a better model -- there's to little information here to judge what the circumstances really are.

  • $\begingroup$ I just spotted a mangled phrase (I broke it during the initial edit), which was quite misleading. Now fixed. $\endgroup$
    – Glen_b
    May 12, 2019 at 2:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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