ARIMA - Additional Forecasting Frequencies

[BS_THE_ANALYST]

Hey all,

Currently, I'm facing a problem where I'd like to select the "Other" option for the Target Field Frequency and provide a different forecasting frequency than what is listed in ARIMA's defaults. Has anyone faced/resolved this issue before?

If you're wondering why isn't the "Other" option satisfactory for me:

my understanding

If you imagine I had 24 rows of data (assume they are arranged chronologically for this explanation). If I selected the Monthly option, the model will know that row1 matches to row13, row2 matches to row14 and so forth as we're telling it what the seasonal periods are. Same situation with Quarterly, it would assume row1 matches to row5, row2 matches to row6 and so forth.

So, without being able to supply a number (expected frequency) in the "Other" option, I can't see how I'll be able to supply it with what the seasonal periods are, so that the model knows what to expect. 

I've tried:
I went into the underlying R code and tried to manually add the option of "13" (which is the forecasting frequency in my case). I did this by overwriting a current option, see the yellow highlighted area below (had no luck with this) and also creating a new radio button and added an additional option that would allow me to select "13" via the interface, you can achieve this by replicating the green highlighted area (this didn't work either).

 

 

Conclusion
That's where I'm up to, no success, unfortunately. Any help would be greatly appreciated.

All the best,
BS

[alexnajm]

I tried on 2023.2 and 2022.3 and I see the same thing, which makes me think this is normal behavior and it relies on the cadence of the data coming into the tool!

[BS_THE_ANALYST]

@alexnajm great, thanks for letting me know.

Just a theoretical example, but let's say we split the year in half, i.e., we have 2 periods. How would we be able to tell the ARIMA model that period 1 won't occur again for another 2 rows in our data? I can only see this being possible with the ability to select an option. For example, here are the actual values 'under-the-hood':

If the data is hourly, we would select hourly, and then it knows the frequency is '24'. For any given row, we will see this period occur again 24 rows from where it's situated. I want to emulate this behavior, but I need to tell the model '13' or '8,' problem-dependent, of course.

Hopefully that makes sense!