Difference between Econometrics ARIMA models and Identification Toolbox AR/ARMAX models

Dear all,

I have a code in which I need to predict the next 50 samples of network traffic, based on the knowledge of the previous 50 samples, based on an ARMA/ARIMA model. Current order are very low (1 for AR and 1 for MA, and not larger than 2).

My current code makes use of the Econometrics toolbox functions arima and estimate to contract and estimate the model, and then simulate it. Actually, the only outcome I am interested with is the sum of the predicted traffic. That is, having the estimate result in 'model', and called y the result of simulate(model,50), I need only sum(y).

I also saw that the identification toolbox has analogous functions for the estimation of model parameters for ARMA/ARIMA models, through functions ar/armax... However, the estimated parameters are quite different!

My question is the following:

Given that I need to introduce the estimation within a loop, the simulation takes more than 1 second per loop with the current implementation, most of the time spent with the arima model estimation (e.g. calls of fmincon, etc...)

• Is there a way to decrease (substantially) this time?
• The use of the system Identification toolbox can be of help?
• More in general, to me the use of the SIT seems more natural than the ET (the problem has nothing to do with econometrics ;) ). Why the parameters are so different, and possibly where am I wrong?

Thank you for all your support.

Best regards

Federico

[ronak feizimirkhani]

Hi Federico, I am writting here because i have the same question. I am trying to estimate the ARMA model over my measured time series and i am only using the Identification Toolbox. You mentioned that you have tried both and the estimated parameters are differente. I want to know if 1. the difference is considerable? 2. the estimated FITs are totaly different or it's just in te precision term? 3. can i trust on the armax estimation parameters?

Thank you so much and sorry that I did not help you but asked questions :-(

[Shaowu Pan]

Same question

For my case, I test varm vs armax, they match.

But for arima vs armax, they differ, a lot. armax is giving better results.