2

I am new to time series, and here i have a question about ARIMA- determine the order of MA and AR.

In one post i learnt that usually we use the ACF and PACF plot to determine whether we should use a AR or MA term. and usually we use only one of the terms while leaving the other term zero (eg. 1,1,0). If the ACF plot shows a gradual fall while PACF shows a sudden cutoff at lag k, then we take will take an AR(k) value, leaving the MA(0). post:http://people.duke.edu/~rnau/411arim3.htm

However, in another post of ARIMA using software STATA, the story is contradicting. The post clearly shows the same scenario, where the ACF plot falls gradually (eg. 6 lags) and the PACF shows a sudden cutoff at lag 1, but the second post instead test out several tentative solutions (1,1,1) (1,1,2) (1,1,3) (1,1,4) (1,1,5) (1,1,6). post:https://www.projectguru.in/publications/univariate-arima-model-time-series-analysis-stata/

The 2 posts are contradictory to me, because the first teach me usually we make use of either one of the AR/MA terms, and use the PACF lag number as the order of ACF (in the AR case). But the second post says we could try out different scenarios, and use ACF lag number as the order of MA and PACF lag number as the order of AR at the same time.

Could someone clarify about this question? I am really confused! Thank you so much! :)

Sincerely.

user7
  • 41
  • 2

1 Answers1

1

The Guru is NOT a GURU , follow the first thread . You might also want to read state-of-art of ARIMA as it details the problem with data that may not be clean.

a good way is to follow https://newonlinecourses.science.psu.edu/stat510/node/49/enter image description here

IrishStat
  • 27,906
  • 5
  • 29
  • 55
  • Hi thank you for your reply. I found a youtube tutorial also on this topic using STATA and the youtuber used the second idea to perform calculations. May I know the reasons? Thank you! https://www.youtube.com/watch?v=Rd67Tin8igA – user7 Feb 22 '19 at 01:25
  • 1
    the trial and error approach fails because you can easily have over modelling for both MA and AR/differencing ,, essentially cancelling each other out. – IrishStat Feb 22 '19 at 01:30
  • https://stats.stackexchange.com/questions/380599/is-it-possible-to-automate-time-series-forecasting/380634#380634 can also be useful. – IrishStat Feb 22 '19 at 02:11