Simulations: Add-factor interaction with ARMA errors

For technical support, tips and tricks, suggestions, or any other information regarding the EViews model object.

Moderators: EViews Gareth, EViews Moderator

CrisisStudent
Posts: 48
Joined: Mon Nov 16, 2020 10:52 am

Simulations: Add-factor interaction with ARMA errors

Postby CrisisStudent » Sun Dec 19, 2021 8:48 am

Hi all,

I understand that in deterministic environment add-factors are treated as exogenous variables, so that their effects are not propagated via ARMA errors - i.e. change in add-factor in given period will not have any effects past given period carried through ARMA errors*.

Does the same apply to stochastic simulations? That is, are the innovation shocks propagated via the ARMA errors in stochastic solves?

In either case, what is the best way to change this and allow add-factor shocks to be propagated through ARMA errors?

Thanks!
CrisisStudent


* Of course this does not apply to lagged dependent variables.

CrisisStudent
Posts: 48
Joined: Mon Nov 16, 2020 10:52 am

Re: Simulations: Add-factor interaction with ARMA errors

Postby CrisisStudent » Thu Jan 06, 2022 9:45 am

Hi guys, did you have time to consider this question?

EViews Matt
EViews Developer
Posts: 560
Joined: Thu Apr 25, 2013 7:48 pm

Re: Simulations: Add-factor interaction with ARMA errors

Postby EViews Matt » Thu Jan 06, 2022 5:27 pm

Hello,

Sorry your post escaped my notice earlier. Unfortunately, add factors are not treated as innovation shocks even in stochastic solves and there are no options for altering this behavior.

CrisisStudent
Posts: 48
Joined: Mon Nov 16, 2020 10:52 am

Re: Simulations: Add-factor interaction with ARMA errors

Postby CrisisStudent » Thu Jan 06, 2022 11:32 pm

Thanks Mat.

Can I brainstorm with you an idea? 8)

We had to develop our own stochastic simulation code in Eviews to give us flexible control over some of the functionalities, which means I could implement a solution to address the issue with ARMA errors propagation. That said, I am not sure how to best address the issue. The only thing that occurred to me was to manually adjust the stochastic innovation series our codes create - i.e. treat the innovations as innovations for ARMA process and then calculate the final innovations as result of the process - which I think could work. But maybe one could somehow adjust the model structure itself to achieve the same result in better way...? Just throwing out there in case you will have some smart idea. :idea:

CrisisStudent

CrisisStudent
Posts: 48
Joined: Mon Nov 16, 2020 10:52 am

Re: Simulations: Add-factor interaction with ARMA errors

Postby CrisisStudent » Fri Jan 07, 2022 1:18 am

Actually, now I re-read your reply and I just wanted to make sure one thing. You said "add factors are not treated as innovation shocks even in stochastic solves". But what about stochastic innovations themselves, do they interact with ARMA errors? (That is: If the stochastic innovations are treated as add-factors, then the answer is clear, but I am not sure if that is the case).

EViews Matt
EViews Developer
Posts: 560
Joined: Thu Apr 25, 2013 7:48 pm

Re: Simulations: Add-factor interaction with ARMA errors

Postby EViews Matt » Tue Jan 11, 2022 6:19 pm

Hello,

Addressing your last point first, stochastic innovations do interact with ARMA errors as you'd expect. Add factors are the oddity. In effect, residuals/innovations are determined before add factors are applied, so add factors don't propagate through ARMA terms.

Your approach for implementing such functionality seems reasonable to me. I have a vague idea that perhaps an ARMA equation could be split into two equations within a model to introduce the control you want, but I don't know if that idea would really pan out or even provide any advantage over what you're already developing. Perhaps your work would make a nice addin? 8)

CrisisStudent
Posts: 48
Joined: Mon Nov 16, 2020 10:52 am

Re: Simulations: Add-factor interaction with ARMA errors

Postby CrisisStudent » Mon Jan 17, 2022 10:23 am

Thanks Matt for your reply, this makes everything clear.

Unfortunately, these codes are developed as part of my formal employment, so can't share them with wider community as an add-in... :roll:


Return to “Models”

Who is online

Users browsing this forum: No registered users and 7 guests