On Thu, 25 Mar 2010, Henrique Andrade wrote:

I would like to suggest some improvements (I think ;-)) to the VAR model window. My aim goal is to make VAR model
window more close to OLS model window in terms of "easiness of use". Additionally, it would be a good change for
Gretl 2.0. Some of the suggestions have been inspired by EViews.

(1) Impulse-response: I think it could be useful if we had more options regarding to the impulses. EViews has the
option "impulse definition" where we can choose the decomposition method (Residual - one unit; residual - one
std. deviation; Cholesky - dof adjusted; Cholesky - no dof adjustment; generalized impulses; structural
decomposition; and user specified.

The most interesting item in this list (to me) is the structural decomposition, which is what my SVAR package does. For the moment, its development is on hold, but a few people have tested it (I asked for help with the testing in a message on the user list about 4 weeks ago: http://lists.wfu.edu/pipermail/gretl-users/2010-March/004483.html) and the basic cases seem to work ok.

Could you please send your SVAR package to me? ;-)

As for the other requests:

Cholesky: dof/no-dof. Do you reaaly think this would be of any use?

There was an e-mail about that about 10 days ago (http://lists.wfu.edu/pipermail/gretl-users/2010-March/004551.html). In the estimated VAR window Gretl shows the standard-error with standard-error with degrees of freedom correction, but in the impulse-response it shows standard-error without degrees of freedom correction. Moreover, sometimes we need to know the impact of different shocks (not only one standard deviation).

Generalised impulses: there was a thread about those about a month ago: http://lists.wfu.edu/pipermail/gretl-users/2010-February/004456.html
User specified: what are they????

(2) Impulse-response and variance decomposition: It would be great if we had the option "Cholesky ordering" (I
know we can choose this before the estimation, but I think this is not so good).

This wouldn't be difficult to do. Basically, all that would be needed would be to interchange rows and columns of Sigma (the VAR residuals vcv matrix) as desired, compute the Cholesky decomposition of that (call it C), and then put the rows and columns of C back in the original order. After that, compute IRFs and FEVDs on the basis of the reordered C. Doable, but (given the complication in the user interface) is it worth it?

I know that all these things could be made by Gretl using scripts, but I really think that these features are important and deserve a bit of the great "Gretl's User Friendliness". ;-)

On Mon, 29 Mar 2010, Riccardo (Jack) Lucchetti wrote:

> On Thu, 25 Mar 2010, Henrique Andrade wrote:
>

> > (2) Impulse-response and variance decomposition: It would be
> > great if we had the option "Cholesky ordering" (I know we can
> > choose this before the estimation, but I think this is not so
> > good).
>
> This wouldn't be difficult to do. Basically, all that would be
> needed would be to interchange rows and columns of Sigma (the
> VAR residuals vcv matrix) as desired, compute the Cholesky
> decomposition of that (call it C), and then put the rows and
> columns of C back in the original order. After that, compute
> IRFs and FEVDs on the basis of the reordered C. Doable, but
> (given the complication in the user interface) is it worth it?

IMO offering on-the-fly reordering is quite a nice convenience to
the user. This is now in CVS, but it needs testing. It will go
into the snapshots shortly.

I'd tested and, in my humble opinion, it looks very good. Thanks Allin!

Best,