3

I am using the table below as model selection tool (at least as starting point)

enter image description here

Let's say that I choose a proper model according to the table and I get nice ACF and PACF out of it, but either my AR term or my MA term is pretty high, is there a way to simplify it?

Note: I don't know if it is relevant, but I am using R.

Richard Hardy
  • 54,375
  • 10
  • 95
  • 219
user158013
  • 51
  • 3
  • .......... see my answer – IrishStat Nov 15 '17 at 12:02
  • 1
    What do you mean by "either my AR term or my MA term is pretty high"? I assume you mean the AR or MA orders? This can happen with seasonality (and then you should be doing seasonal differencing), but otherwise [very rarely](https://stats.stackexchange.com/q/285093/1352). – Stephan Kolassa Nov 15 '17 at 12:04
  • Or perhaps I actually misunderstood the question... Could you clarify? – Firebug Nov 15 '17 at 12:27
  • My question is, if after an iterative process I find a proper model but the AR or MA order (or both) is pretty high, should I consider a way to simply them? If yes, how should I procede? – user158013 Nov 15 '17 at 12:34
  • 1
    Could you try to make your title more specific? And also explain what you mean by "AR term or my MA term is pretty high"? – Richard Hardy Nov 15 '17 at 12:37

1 Answers1

0

keep in mind that there is a presumption of no pulses, no level shifts , no seasonal pulses and no local time trends in your approach. If you treat the identification in a holistic manner , the ARIMA structure is often quite simplified and more correct. Additional assumptions are that both the parameters and the model error variance are constant over time. One has to read/understand the fine print.

IrishStat
  • 27,906
  • 5
  • 29
  • 55
  • 1
    I am not sure I totally understood your answer: but my question is not actually about the selection method; it's more about wheter or not I should take for good a model with high AR or MA order or if I should argue them somehow. – user158013 Nov 15 '17 at 12:34
  • i would argue them by starting simple and proceed with an iterative self-checking process to build up the model to a point of sufficiency. The software/approach you are using to do model identification is probably way over-parameterizing. If you wish to post a data set and your model I will try and help further. – IrishStat Nov 15 '17 at 12:41