8:46 p.m.
Hi,
now that I have my first gmm estimates produced by gretl, I would like
to know what else is possible -- I already found out that "restrict"
apparently is _im_possible. What about saving the estimated coefficients
or covariance matrix for further (scripting) use? (I have read the gmm
chapter but couldn't find anything.)
thanks,
sven
11:49 p.m.
On Fri, 21 Sep 2007, Sven Schreiber wrote:
now that I have my first gmm estimates produced by gretl, I
would like to know what else is possible -- I already found out
that "restrict" apparently is _im_possible. What about saving
the estimated coefficients or covariance matrix for further
(scripting) use? (I have read the gmm chapter but couldn't find
anything.)
You already have the coefficients, since you defined them in the
first place (and they are modified in the course of estimation).
The covariance matrix is obtained in the usual way:
matrix V = $vcv
The usual "restrict" apparatus can't apply, because we can't
assume that the parameters enter the model linearly.
You should, however, be able to save the value of the GMM
criterion function, so you can cook up your own restriction test
-- but if I'm not mistaken, you can't at present.
Allin.
4:13 p.m.
Allin Cottrell schrieb:
On Fri, 21 Sep 2007, Sven Schreiber wrote:
> now that I have my first gmm estimates produced by gretl, I
> would like to know what else is possible -- I already found out
> that "restrict" apparently is _im_possible. What about saving
> the estimated coefficients or covariance matrix for further
> (scripting) use? (I have read the gmm chapter but couldn't find
> anything.)
You already have the coefficients, since you defined them in the
first place (and they are modified in the course of estimation).
Right, so that was sort of a stupid question...
The covariance matrix is obtained in the usual way:
matrix V = $vcv
ok good to know; actually I just properly realized a minute ago that GMM
is also available in a dialog window, not just via scripts... let's see
what that gives...
The usual "restrict" apparatus can't apply, because we can't
assume that the parameters enter the model linearly.
You should, however, be able to save the value of the GMM
criterion function, so you can cook up your own restriction test
-- but if I'm not mistaken, you can't at present.
yes that would be nice.
-sven
10:17 a.m.
New subject: bugs&wishlist (was Re: gmm usage)
Ok, here are some issues with the GMM dialog:
First what I view as bugs:
1) choosing the "iterated" option doesn't seem to be honored, I get
1-step output.
2) the convergence tolerance setting is shown in the GMM output (good),
but not anymore after saving to the session window and reopening (bad).
3) unrelated to GMM: horizontally resizing the session window leads to
icons dis- and/or reappearing
And some small wishes relating to the GMM GUI:
4) An option to set the tolerance would be nice (simple entry field or
even drop-down list with a range of digits maybe)
5) A checkbox for requesting the $vcv; maybe also an entry field to
assign a name to the matrix that would be created in the session.
And finally, for Gmm in general (script and GUI use):
6) Ability to specify standard options for the (initial) weighting
matrix. AFAICS that would be a choice of two: the identity matrix or the
inverse covariance matrix of the instruments. Of course over-rideable by
an explicit weights command.
thanks,
sven
Allin Cottrell schrieb:
On Fri, 21 Sep 2007, Sven Schreiber wrote:
> now that I have my first gmm estimates produced by gretl, I
> would like to know what else is possible -- I already found out
> that "restrict" apparently is _im_possible. What about saving
> the estimated coefficients or covariance matrix for further
> (scripting) use? (I have read the gmm chapter but couldn't find
> anything.)
You already have the coefficients, since you defined them in the
first place (and they are modified in the course of estimation).
The covariance matrix is obtained in the usual way:
matrix V = $vcv
The usual "restrict" apparatus can't apply, because we can't
assume that the parameters enter the model linearly.
You should, however, be able to save the value of the GMM
criterion function, so you can cook up your own restriction test
-- but if I'm not mistaken, you can't at present.
Allin.
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
2:02 a.m.
New subject: bugs&wishlist (was Re: gmm usage)
On Mon, 24 Sep 2007, Sven Schreiber wrote:
Ok, here are some issues with the GMM dialog:
First what I view as bugs:
1) choosing the "iterated" option doesn't seem to be honored, I get
1-step output.
I believe that was a "cosmetic" issue: the iteration option was in
fact honored, but if the thing converged immediately the printout
said "one-step". I think this should now be fixed.
2) the convergence tolerance setting is shown in the GMM output
(good),
but not anymore after saving to the session window and reopening (bad).
OK, will think about that.
3) unrelated to GMM: horizontally resizing the session window leads
to
icons dis- and/or reappearing
Hmm. Anyone else seen that? I tried on my own machine and
couldn't replicate the problem. It sounds like a GTK issue: is it
specific to Windows?
And some small wishes relating to the GMM GUI:
4) An option to set the tolerance would be nice (simple entry field or
even drop-down list with a range of digits maybe)
5) A checkbox for requesting the $vcv; maybe also an entry field to
assign a name to the matrix that would be created in the session.
And finally, for Gmm in general (script and GUI use):
6) Ability to specify standard options for the (initial) weighting
matrix. AFAICS that would be a choice of two: the identity matrix or the
inverse covariance matrix of the instruments. Of course over-rideable by
an explicit weights command.
These all seem like desirable things. It's just a small matter of
how many hours there are in the day.
Allin.
7:50 a.m.
New subject: bugs&wishlist (was Re: gmm usage)
Allin Cottrell schrieb:
On Mon, 24 Sep 2007, Sven Schreiber wrote:
> 3) unrelated to GMM: horizontally resizing the session window
leads to
> icons dis- and/or reappearing
Hmm. Anyone else seen that? I tried on my own machine and
couldn't replicate the problem. It sounds like a GTK issue: is it
specific to Windows?
probably, but haven't checked
> And some small wishes relating to the GMM GUI:
>
> 4) An option to set the tolerance would be nice (simple entry field or
> even drop-down list with a range of digits maybe)
>
> 5) A checkbox for requesting the $vcv; maybe also an entry field to
> assign a name to the matrix that would be created in the session.
>
> And finally, for Gmm in general (script and GUI use):
>
> 6) Ability to specify standard options for the (initial) weighting
> matrix. AFAICS that would be a choice of two: the identity matrix or the
> inverse covariance matrix of the instruments. Of course over-rideable by
> an explicit weights command.
These all seem like desirable things. It's just a small matter of
how many hours there are in the day.
Of course! This is just feedback I give whenever I explore something in
gretl that I haven't used before.
-sven
6049
days inactive
6055
days old
5 comments
2 participants
participants (2)
-
Allin Cottrell -
Sven Schreiber