Page 1 of 1

actexist

Posted: Thu Aug 20, 2020 12:39 am
by tvonbrasch
Hi

We simulate our model using the actexist option in the exclude command. We would like to generate add-factors so that we can turn the actexist option off and the re-simulate the model and still get the identical path we had when the actexist option was activated. I thought these commands would do it

Code: Select all

m.exclude(actexist=t)  {%exclude}   
m.solve
m.exclude(actexist=f)  {%exclude}   
m.addinit(v=a) @all


but, if I then solve the model again (m.solve) it does not reproduce the original path. How can I generate add-factors so that the model will reproduce the path when the actexist option was active?
t

Re: actexist

Posted: Fri Aug 21, 2020 9:57 am
by EViews Matt
Hello,

I'm also surprised that didn't work as expected. However, in this case I believe the (v=n) option to addinit is a workable substitute.

Re: actexist

Posted: Fri Aug 21, 2020 10:46 pm
by tvonbrasch
Hi

Thanks for your reply, Matt. It does not work with v=n. I am probably doing something wrong. Can you please take a look at the following code:

Code: Select all

'set smpl
   smpl 2018 2025
'turn on actexist
   {%m}.exclude(actexist=t) {%excludes}
'solve model
   {%m}.solve
'copy solution
   copy *_kt *_kt2
'set residuals to match actives
   {%m}.addinit(v=a) @all  '(it does not work with v=n)
'turn off actexist
   {%m}.exclude(actexist=f) {%excludes}
'solve the model again
   {%m}.solve
'compare solutions
   smpl 2019:3  2020
   line ur_kt ur_kt2


This produces the following graph:
actexist_forum_comparing_solutions.PNG
actexist_forum_comparing_solutions.PNG (80.36 KiB) Viewed 1712 times


As you can see, the simulation does not reproduce the actuals-value in 2020:2.

Is there an obvious fix that I should do? If not, I can share the workfile with you on monday.
t

Re: actexist

Posted: Sun Aug 23, 2020 12:36 pm
by EViews Matt
Hmm. Yes, please do share your workfile.

Re: actexist

Posted: Tue Aug 25, 2020 5:49 pm
by EViews Matt
I believe the issue is a set of missing add factors for 10 variables named pcja??_a. Some series with that name pattern are add factors for variables named pcja??, but the 10 variables I speak of unfortunately are named like add factors and would have their own add factor series named pcja??_a_a. Simply reenabling add factors for all of the model's variable appears to fix the issue, e.g., inserting "{%m}.addassign @all" before the addinit statement.

Re: actexist

Posted: Wed Aug 26, 2020 3:58 am
by tvonbrasch
Hi

Thanks for looking into this, Matt. The model object did have add-factors, also for those equations, but they were labelled with the alias "_a2". I changed the add-factors based on the forum post: http://forums.eviews.com/viewtopic.php?f=10&t=20252. As you stated in that post, the proposed solution "doesn't appear to break anything, but no guarantees." I guess this example illustrates that EViews cannot handle that add-factors are given another alias than "_a" ? Or is there a way to overcome this that I have not thought of yet?

Thanks again Matt for looking into this
t

Re: actexist

Posted: Wed Aug 26, 2020 4:27 pm
by EViews Matt
In the model you sent me, the last 10 equations did not have add factor enabled. The odd add factor names worked fine for me, e.g., variable pcja00's add factor is named pcja00_a2 because pcja00_a is another variable in the model. I think the root issue was just that 10 variables were missing add factors, e.g., pcja00_a didn't have an add factor (to be named pcja00_a_a), which turned out to be necessary to match the trajectory you wanted.

Re: actexist

Posted: Thu Aug 27, 2020 9:11 am
by tvonbrasch
ah, right, now I understand. Thanks Matt!
t