Long run restrictions for a structural VAR
Moderators: EViews Gareth, EViews Moderator
Long run restrictions for a structural VAR
Dear Colleagues,
I have estimated a structural VAR model and I am imposing short run and long restrctions on the model using eviews. However, I am having difficulties with estimating the long run restrictions because eviews seems to impose a lower triangular matrix on the long run restriction. Because my long run restricted model allows some coefficients to be estimated in the upper triangular matrix, I am getting an error message saying "the structural VAR objective function cannot be evaluated at the initial parameter values". I want to ask if anyone have an idea as to how I can overcome this problem?
Thank you
I have estimated a structural VAR model and I am imposing short run and long restrctions on the model using eviews. However, I am having difficulties with estimating the long run restrictions because eviews seems to impose a lower triangular matrix on the long run restriction. Because my long run restricted model allows some coefficients to be estimated in the upper triangular matrix, I am getting an error message saying "the structural VAR objective function cannot be evaluated at the initial parameter values". I want to ask if anyone have an idea as to how I can overcome this problem?
Thank you

 EViews Developer
 Posts: 239
 Joined: Thu Apr 25, 2013 7:48 pm
Re: Long run restrictions for a structural VAR
Hello,
There's no implicit restriction form on the longrun matrix, you can restrict any element(s) you wish as long as you completely identify the system. That error can occur when, for example, the choice of initial values for the estimation happens to produce a singular matrix or other numerically troublesome result. Have you tried starting the optimization with a different/random set of initial values?
There's no implicit restriction form on the longrun matrix, you can restrict any element(s) you wish as long as you completely identify the system. That error can occur when, for example, the choice of initial values for the estimation happens to produce a singular matrix or other numerically troublesome result. Have you tried starting the optimization with a different/random set of initial values?
Re: Long run restrictions for a structural VAR
Thank you for the reply. You asked if I tried starting the optimization using different initial values. The issue is that I am not sure what the initial values are, and this may be the problem. So I want ask what exactly are the initial values? You also mentioned that there are no implicit restrictions in the long run matrix as long as the system is completely identified. Are you referring to the identification of the short run matrix or the long run matrix? The identification scheme used to identify both matrices is based on theory and previous literature, and the results show that the model is just identified. Am not sure if there is something else that I am not doing right.

 EViews Developer
 Posts: 239
 Joined: Thu Apr 25, 2013 7:48 pm
Re: Long run restrictions for a structural VAR
Lets start with the initial value issue, since that's easy to fix (if it's the cause of your difficulty). When estimating an SVAR, EViews uses an initial value of .1 for all estimates by default. If you open your VAR, go to Proc > Estimate Structural Factorization, and switch to the Optimization Control tab, I assume the Starting Values option is set to Fixed: 0.1. Try reestimating your SVAR with one of the Drawn from... options instead. Those options assign random values to the initial estimates, which may avoid numerical special cases triggered by all .1's. If you're running your SVAR estimate from the command window or a program instead of the GUI, you can add the f0=u or f0=n option to the svar proc.
Re: Long run restrictions for a structural VAR
Thank you once again for the guidance and suggestions. I have tried the options you suggested from the Drawn from... and when I estimate the long run F matrix with some coefficients in the upper diagonal, I still get an error message saying "Optimization may be unreliable (first or second order conditions not met)". What I noticed however, is that the long run impulse response matrix F has a bracket where it states 'triangular'. To be specific, the long run matrix is stated as "Recursive longrun impulse response (F triangular)". The other matrices such as the A and S matrices have no any such restrictive definition and with those matrices, it is easy to estimate coefficients on either side of the diagonal. I am therefore, wondering whether the long run matrix is just programmed to be lower triangular such that all coefficients in the upper diagonal and cannot be estimated (I may be wrong, but this is just my little thought).

 EViews Developer
 Posts: 239
 Joined: Thu Apr 25, 2013 7:48 pm
Re: Long run restrictions for a structural VAR
Those "restriction presets" are just simple, common sets of restrictions that can be customized further. You're not limited by those presets, e.g., the F matrix doesn't need to be triangular. The new error message you're receiving could be caused by many things, such as bad initial values or poor identification of the SVAR model. Reestimating with different initial values may produce better results if the former is true, while changing the restrictions you're imposing can address the latter.
SVAR Matrices
Hello,
I'm working on SVAR to estimate my model, I found this eviews video (note below link) very helpful however I need to know how to determine MatrixA and MatrixB as mentioned in this video. I have 1 dependent variable and 6 independent variables.
https://www.youtube.com/watch?v=JFKZn665D8c&t=51s
Thank you very
I'm working on SVAR to estimate my model, I found this eviews video (note below link) very helpful however I need to know how to determine MatrixA and MatrixB as mentioned in this video. I have 1 dependent variable and 6 independent variables.
https://www.youtube.com/watch?v=JFKZn665D8c&t=51s
Thank you very

 EViews Developer
 Posts: 239
 Joined: Thu Apr 25, 2013 7:48 pm
Re: Long run restrictions for a structural VAR
Hello,
Are you asking about how you should restrict those matrices? There is no one correct way to do so, the restrictions are part of the assumptions of your model. It's up to you to specify how you believe the variables in your model are related to one another.
Are you asking about how you should restrict those matrices? There is no one correct way to do so, the restrictions are part of the assumptions of your model. It's up to you to specify how you believe the variables in your model are related to one another.
Re: Long run restrictions for a structural VAR
Hi,
I am having an issue with a very similar topic. Eviews does not appear to be implementing my restrictions on the longrun F matrix, or S matrix. I have tried specifying it a number of ways (point & click and programming) and it still does not work.
That is, the restrictions are stated at the top of the output as I intended. But when I look at the estimated S and F matrices at the bottom of the output, my restrictions have not been implemented. Instead S is always lower triangular and F always has estimates for all elements.
Any advice would be grand.
Eviews 10 Enterprise Edition build 27 July 2017
Alex
I am having an issue with a very similar topic. Eviews does not appear to be implementing my restrictions on the longrun F matrix, or S matrix. I have tried specifying it a number of ways (point & click and programming) and it still does not work.
That is, the restrictions are stated at the top of the output as I intended. But when I look at the estimated S and F matrices at the bottom of the output, my restrictions have not been implemented. Instead S is always lower triangular and F always has estimates for all elements.
Any advice would be grand.
Eviews 10 Enterprise Edition build 27 July 2017
Alex

 EViews Developer
 Posts: 239
 Joined: Thu Apr 25, 2013 7:48 pm
Re: Long run restrictions for a structural VAR
Hello,
If you have restrictions on A and/or B, it's possible the optimizer may settle on a solution that violates S and/or F restrictions. There have been some additional warning and error messages added to the SVAR system since your build, so updating EViews and retrying the estimation may yield more information. Posting the example that is failing, if possible, would also be helpful.
If you have restrictions on A and/or B, it's possible the optimizer may settle on a solution that violates S and/or F restrictions. There have been some additional warning and error messages added to the SVAR system since your build, so updating EViews and retrying the estimation may yield more information. Posting the example that is failing, if possible, would also be helpful.
Last edited by EViews Matt on Fri May 11, 2018 9:44 am, edited 1 time in total.
Re: Long run restrictions for a structural VAR
Thanks for the reply.
As I am using a build provided by my university, it may be difficult to get them to update it.
The workfile is attached, along with a screen shot of the output.
The code is below.
As I am using a build provided by my university, it may be difficult to get them to update it.
The workfile is attached, along with a screen shot of the output.
The code is below.
Code: Select all
var q3svardiff.ls 1 2 dprod de du ec
matrix(4,4) a_chol
a_chol.fill 1, NA, NA, NA, 0, 1, NA, NA, 0, 0, 1, NA, 0, 0, 0, 1
matrix(4,4) b_chol
b_chol.fill NA, 0, 0, 0, 0, NA, 0, 0, 0, 0, NA, 0, 0, 0, 0, NA
matrix(4,4) fmat
fmat.fill NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, 0, NA, NA, NA
matrix(4,4) smat
smat.fill NA, 0, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA
q3svardiff.svar(s=smat,f=fmat,a=a_chol,b=b_chol)
 Attachments

 lrres.png (41.74 KiB) Viewed 279 times

 lrtest.wf1
 (53.74 KiB) Downloaded 10 times

 EViews Developer
 Posts: 239
 Joined: Thu Apr 25, 2013 7:48 pm
Re: Long run restrictions for a structural VAR
That S is lower triangular is a consequence of A being lower triangular and B being diagonal (remember that S = A^1 * B). As I suspected, running your SVAR with an uptodate version of EViews generates an error indicating that the optimization converged before the S/F restrictions were satisfied. Reestimating with different initial values multiple times produces the same (bad) results, which hints that the restrictions may be misspecified. I then took a look at the analytical consequences of your restrictions:
 S(2,1) = A(2,1) * B(1,1) = C(1) * C(7) = 0. Clearly, this is only satisfied if C(1) = 0 (under the assumption that the diagonal of B cannot contain zeros). This is likely problematic since the model is already overspecified and these extra restrictions don't match the behavior of the data.
 F(1,4) = Psi(1,4) * B(4,4) = Psi(1,4) * C(10) = 0. Unless Psi(1,4), which is based on the underlying VAR estimates, happens to be zero, this restriction is impossible to satisfy (again under the assumption that the diagonal of B cannot contain zeros).
Last edited by EViews Matt on Sun May 13, 2018 9:55 am, edited 3 times in total.
Re: Long run restrictions for a structural VAR
Fantastic, thanks for the help.
I will go over my identification restrictions more thoroughly and hopefully get some traction on it.
Still very useful to know that my build behaves in this way.
Alex
I will go over my identification restrictions more thoroughly and hopefully get some traction on it.
Still very useful to know that my build behaves in this way.
Alex

 EViews Developer
 Posts: 239
 Joined: Thu Apr 25, 2013 7:48 pm
Re: Long run restrictions for a structural VAR
Updated my analysis above. Something about it was bothering me, until I finally realized I had made a typo.
Who is online
Users browsing this forum: No registered users and 3 guests