Page **1** of **3**

### Imposing Restrictions on SVAR

Posted: **Sat Jun 13, 2009 7:56 pm**

by **marco54**

Hi, there.

I have a question about imposing both short-run restriction and long-run restriction on SVAR. The problem is as follows:

In an experiment, I estimate a 5 variables VAR(6), and I want to use the Proc/Estimate Structural Factorization to identify an SVAR. I know that if the model can be identified, it'll need 2k^2-k(k+1)/2=35 restrictions.

I make the short-run restriction as follows:

@e1=c(1)*@u1

@e2=c(2)*@u1+c(3)*@u2

@e3=c(4)*@u1+c(5)*@u2+c(6)*@u3

@e4=c(7)*@u1+c(8)*@u2+c(9)*@u3+c(10)*@u4+c(11)*@u5

@e5=c(12)*@u1+c(13)*@u2+c(14)*@u3+c(15)*@u4+c(16)*@u5

And I found that it still need another restriction to meet the requirement of just identifiable. According to the economic theory, I impose an long-run restriction like @lr4(@u5)=0.

But the problem comes. When I write all the restrictions above in the "text" box, and press yes. It informed me that "Internal Error 500".

How can I do to impose both long-run and short-run restriction on the VAR(p) to make the SVAR model and get the IRF?

Muchas Gracias!

### Re: Imposing Restrictions on SVAR

Posted: **Sat Jun 13, 2009 10:52 pm**

by **startz**

It may help to say which version you are using. And do you have the most recent updates?

### Re: Imposing Restrictions on SVAR

Posted: **Sat Jun 13, 2009 10:57 pm**

by **marco54**

The version I use is EViews6 Enterprise Edition and the update version is Apr 4 2008 Build

### Re: Imposing Restrictions on SVAR

Posted: **Sun Jun 14, 2009 10:28 am**

by **startz**

marco54 wrote:The version I use is EViews6 Enterprise Edition and the update version is Apr 4 2008 Build

Try updating from the EViews website. (If that doesn't work, you'll probably have to wait for QMS to open on Monday.)

### Re: Imposing Restrictions on SVAR

Posted: **Tue Feb 09, 2010 8:06 am**

by **pbaranowski**

I have similar problem. I have tried to impose:

@lr2(@u2)=0

@e1=c(1)*@u1

@e2=c(2)*@u2+c(4)*@u1

@e3=c(5)*@u3+c(6)*@u2+c(7)*@u1

- the system is over-identyfied

and it doesn't work. I got "Syntax error. Cannot mix SR and LR restrictions".

When I impose just-identyfying restrictions:

@e1=c(1)*@u1

@e2=c(2)*@u2+c(4)*@u1+c(8)*@u3

@e3=c(5)*@u3+c(6)*@u2+c(7)*@u1

@lr2(@u2)=0

I got "Internal error 500".

I use EViews 6 SE, built 10 Jul 2008

### Re: Imposing Restrictions on SVAR

Posted: **Tue Feb 09, 2010 8:54 am**

by **EViews Gareth**

The answer is the same - you should try updating.

### Re: Imposing Restrictions on SVAR

Posted: **Wed Feb 17, 2010 4:11 am**

by **pbaranowski**

QMS Gareth wrote:The answer is the same - you should try updating.

I have updated EViews (built 10 Jan 2010 now) and the problem still persists.

### Re: Imposing Restrictions on SVAR

Posted: **Thu Jun 22, 2017 2:23 pm**

by **fredmancr**

Hello, I think by the time Eviews cannot manage both SR and LR restrictions in the same SVAR.

### Re: Imposing Restrictions on SVAR

Posted: **Thu Jun 22, 2017 2:33 pm**

by **EViews Gareth**

EViews 10 can handle both LR and SR restrictions.

https://www.youtube.com/watch?v=_nGkJzDlPY4

### Re: Imposing Restrictions on SVAR

Posted: **Wed Jul 19, 2017 1:37 am**

by **nasa**

Hi, i am using Eviews 10 for Svar and once I specify A and B matrix, it tells me that an error: maximum iterations are exceeded.But it works correctly the Svar in Eviews 7. How it emerges such a problem in Eviews 10.

Thanks.

### Re: Imposing Restrictions on SVAR

Posted: **Wed Jul 19, 2017 9:24 am**

by **EViews Matt**

Hello,

If it's truly a problem with slow convergence, the obvious fix is to increase the maximum number of iterations. If that doesn't resolve the issue, please post the workfile/program so we can look at it in more detail. As to why this is now occurring in EViews 10...

The EViews 10 SVAR system uses the improved optimization engine that was introduced in EViews 9. While results should be comparable or better in most cases, any particular estimation could have difficulties. Along with the change in optimization engine a few other "quirks" were fixed. Would I be right in guessing that your SVAR model in EViews 7 had its starting values option set to the default of "From Residual Correlation"? That option no longer exists in EViews 10 (the default is now fixed values of .1). When using the "From Residual Correlation" option, two sequential optimizations were performed, which effectively doubled the maximum number of iterations. Consequently, an SVAR optimization in EViews 10 can appear to require more iterations to converge than in previous versions of EViews, but that's only because versions previous weren't always tracking/reporting all the iterations that were actually performed.

### Re: Imposing Restrictions on SVAR

Posted: **Thu Jul 20, 2017 4:45 am**

by **nasa**

Please use the following endogenous variables from the workfile:

gdppc,dcpi,interest_rate,ner,spendpc and taxpc1. Follow this order as I construct my A and B matrix on the workfile in this order.

exogenous: c,@trend, dummya and dummyb.

One thing,I am using the same starting values fixed 0.1 both in Eviews 7(does not have even that option residual autocorrelation in this version but I can see this option rather in Eviews 9) and Eviews 10.But as I said it is working well with out problem in Eviews 7 but not in Eviews 10 with fixed as a default option.What is wrong with this latest version? Increasing the number of iterations is not working as well.

Thanks.

### Re: Imposing Restrictions on SVAR

Posted: **Thu Jul 20, 2017 4:54 am**

by **nasa**

and the lag number is 1 .

### Re: Imposing Restrictions on SVAR

Posted: **Thu Jul 20, 2017 1:49 pm**

by **EViews Matt**

It appears that your SVAR model is very sensitive to the optimization starting values. In EViews 7, the default values of .1 work well for the optimizer, but not so with the new optimizer in EViews 10. However, drawing the starting values from the standard uniform distribution in EViews 10 seems to permit convergence most of the time. This sensitivity can also be demonstrated in EViews 7, where I've observed that drawing the starting values from the standard normal distribution prevents convergence much of the time.

### Re: Imposing Restrictions on SVAR

Posted: **Fri Jul 21, 2017 12:35 am**

by **nasa**

Ok,thanks

But, what is our base for choosing the starting values from the options ,for example,whether to select the uniform or standard normal or the other options.

Best