Question about VAR object

93 views
Skip to first unread message

Slava Grigoras

unread,
Oct 11, 2011, 6:56:58 AM10/11/11
to iris-t...@googlegroups.com
Hello.
I would like to know if it is possible to estimate a VAR object with exogenious variables, and how can you obtain the value of the log likelihood function of a VAR, directly from the VAR object.
Thank you.
--
Slava Grigoras

Jaromir Benes

unread,
Oct 11, 2011, 9:41:28 AM10/11/11
to iris-t...@googlegroups.com
Hi Slava:

at the moment, you can't have exogenous variables. As a quick fix, you can estimate a VAR with those variables included, and impose zero restrictions on the respective elements of the transtion matrix A (to make the dynamics of the "exogenous" variables independent of the truly "endogenous" ones).

But VARs with exogenous variables are definitely on the list of future extensions.

As for log lik, there is currently only a loglik ratio test between two VARs available, or the various info criteria...

M.

Anita Yang

unread,
Feb 28, 2013, 3:23:30 PM2/28/13
to iris-t...@googlegroups.com
Hi Jaromir,
 
Could you please explain how to impose zero restrictions on the respective elements of the transition matrix A in more details?
Using your tutorial example, now I treat the third variable (ie. Y) as the exogenous variable to a 2nd-order VAR using two variables, R and PI. I have no idea what I should write after 'A=' in the estimate commend. Is the restriction expressed as a numerical value or a matrix? How should I relate it to the corresponding A matrix.

v = VAR();

[v,vdata] = estimate(v,d,{'R','PI','Y'},Inf,'order=',2,'A=',???)
 
 
Thanks a lot.
Anita

Jaromir Benes

unread,
Feb 28, 2013, 3:31:08 PM2/28/13
to iris-t...@googlegroups.com
Hi Anita

You need to create an NY-by-NY-by-P matrix, where NY is the number of variables in the VAR, and P is the order of the VAR (this is a transition matrix), and enter either an NaN for all coefficients that are to be freely estimated, or a particular number for those that are to be fixed.

In your example, let's say you want to restrict the coefficients A(1,2,1) and A(1,2,2) to zero, i.e. the effect of PI{-1} and PI{-2} on R.

You create the A matrix as follows

A = nan(3,3,2);
A(1,2,1) = 0;
A(1,2,2) = 0;

and call the estimate command as follows

[v,vdata] = estimate(v,d,{'R','PI','Y'},Inf,'order=',2,'A=',A);

Let me know whether this is clear enough...

Best
Jaromir

Anita Yang

unread,
Feb 28, 2013, 10:53:16 PM2/28/13
to iris-t...@googlegroups.com
Hello Jaromir,
 
I really appreciate your prompt reply. It seems clear to me now. I am pretty new to the IRIS-toolbox so I am afraid that I might be asking some basic questions you have actually answered before.
I would like to estimate a SVAR with cross-country data. As far as I understand your tutorial example, you only consider one time series data set. What should I do, say, if I want to pool data from 1990Q1 to 2004 Q4 of country A and of country B to do my estimation? (Which program file should I have read so I could see how to do modification?)
 
Best regards.
Anita  

Jaromir Benes

unread,
Mar 1, 2013, 9:10:42 AM3/1/13
to Anita Yang, the-iris-toolbox [discussion forum]
Don't worry :)

What exactly do you have in mind regarding the multiple country estimation exercise? Is it a panel-data kind of regression, or a hierarchical model, or simply combining two data sets? IRIS currently does not support any of those, but it'd be relatively easy to extend the functions to cover the last case (simply combining multiple data sets).

Let me know.

Best,
Jaromir


--
You received this message because you are subscribed to the Google Groups "iris-toolbox-discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to iris-toolbox...@googlegroups.com.
To post to this group, send email to iris-t...@googlegroups.com.
Visit this group at http://groups.google.com/group/iris-toolbox?hl=en.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

Anita Yang

unread,
Mar 1, 2013, 2:24:03 PM3/1/13
to iris-t...@googlegroups.com, Anita Yang
Hello Jaromir,

What I have in mind is to use data from multiple countries and combine them together to do estimation. More specifically, I obtain time series data from eight countries between 2000Q1 to 2010Q4 and then I want to run a reduced-form VAR system by OLS including country dummies. Do you think it's possible to extend the functions to cover this case?

Thanks.
Anita

Jaromir Benes

unread,
Mar 1, 2013, 2:32:31 PM3/1/13
to Anita Yang, the-iris-toolbox [discussion forum]
Possible, it is :) But I'm now rather busy, so can't really promise any delivery date.

Could you please create a feature request on the IRIS web site (go to the Issues section, create a New Issue, and select Feature Request, describing briefly what you want to do).

Best,
Jaromir

Anita Yang

unread,
Mar 1, 2013, 2:46:49 PM3/1/13
to iris-t...@googlegroups.com
I will do that. Thank you so much for all your help.

Anita
Reply all
Reply to author
Forward
0 new messages