ARIMA and ETS input

[coolkidscandie]

Hi,

I started working with ARIMA and ETS and I know that we can either specify components ourselves or leave it for the model (auto). My question is - should we try to pick the components ourselves or leave them on auto and let the model do the work? 

If ARIMA asks for d - do we need to provide data after differencing or does model do differencing itself? If we need to do it - do we start with seasonal or nonseasonal differencing?  

[TrevorS]

Hello @coolkidscandie ,

Thanks for reaching out to the Community! It looks like some additional information could be used to try and get a response here.

Could you please include a copy of your workflow as well as some more information on what you are trying to accomplish? This will help our users to be able to provide you with more feedback!

Thanks!

TrevorS

[HossC]

Hello @coolkidscandie,

This depends on how comfortable you are with time series modeling. With regards to the ARIMA tool, if you are experienced enough to interpret the ACFs and PACFs (Summary of rules for identifying ARIMA models, Identifying the numbers of AR or MA terms in an ARIMA model, Identifying the order of differencing in an ARIMA model), you could consider specifying the components yourself. If not, then it would probably be best to allow the auto functionality to specify the model. This option should also identify the optimal degree of differencing as well, so no need to specify.

If you would like to be as thorough as possible, you can customize the auto model creation parameters. However, this can significantly increase the run-time of your workflow (especially if you select the full enumeration option):

With regards to ETS tool, if you can identify whether or not your time series data represents an additive model or a multiplicative model, you could specify. If not, you will want to allow the auto options to specify the model.

In my experience, the auto options for both the ARIMA tool and the ETS tool do a reasonable job of identifying the correct functional form of the model. Also, I’ve found that Forecasting: Principles and Practice provides a very approachable overview of these modeling types, so it may be worth reviewing this text.