3:11 p.m.
Dear gretl users,
I want to run a Wald test on long-run parameter symmetry based on an
ARDL model. For this I can apply gretl's internal function which is
documented in the help file. However, to be a bit more flexible I
programmed a function which does exactly what is expected, yielding the
same results as gretl's internal procedure. But this does,
unfortunately, hold not in every case.
In the attached script I test the three following hypotheses separately:
1. b[Pinc]/[rho]-b[Ninc]/[rho]=0
2. b[Prt]/[rho]-b[Nrt]/[rho]=0
3. b[Pown]/[rho]-b[Nown]/[rho]=0
For the first two I obtain exactly the same results using either gretl's
internal function or mine. However, for the third one I obtain an error
as the standard errors of the non-linear long-run coefficients
lrse = sqrt(qform(fb,vcv))
cannot be computed. I don't find the issue. Maybe someone has a solution
to this.
Much appreciated.
Artur
3:57 p.m.
On Mon, 1 Sep 2014, Artur T. wrote:
For the first two I obtain exactly the same results using either
gretl's
internal function or mine. However, for the third one I obtain an error
as the standard errors of the non-linear long-run coefficients
lrse = sqrt(qform(fb,vcv))
cannot be computed. I don't find the issue. Maybe someone has a solution
to this.
The result of qform(fb,vcv) may turn out to be a matrix if fb has more
than 1 row. Such a matrix ought to be positive definite, pathological
cases aside, but there may be negative entries off the diagonal. In this
case, sqrt() returns an error, which triggers all the rest of your
problems.
HTH,
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
6:54 p.m.
Thanks for the reply, Riccardo. I am aware of this, but I am a bit
puzzled why it works using gretl's restrict command for this specific
case I described (restricton three). I am wondered why my function
yields the same results for the first two restrictions but not for the
third one. How does gretl exactly compute the Delta method in this case?
Artur
Am 10.09.2014 um 17:57 schrieb Riccardo (Jack) Lucchetti:
On Mon, 1 Sep 2014, Artur T. wrote:
> For the first two I obtain exactly the same results using either gretl's
> internal function or mine. However, for the third one I obtain an error
> as the standard errors of the non-linear long-run coefficients
>
> lrse = sqrt(qform(fb,vcv))
>
> cannot be computed. I don't find the issue. Maybe someone has a solution
> to this.
The result of qform(fb,vcv) may turn out to be a matrix if fb has more
than 1 row. Such a matrix ought to be positive definite, pathological
cases aside, but there may be negative entries off the diagonal. In this
case, sqrt() returns an error, which triggers all the rest of your
problems.
HTH,
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
7:25 p.m.
On Wed, 10 Sep 2014, Artur T. wrote:
Thanks for the reply, Riccardo. I am aware of this, but I am a bit
puzzled why it works using gretl's restrict command for this specific
case I described (restricton three). I am wondered why my function
yields the same results for the first two restrictions but not for the
third one. How does gretl exactly compute the Delta method in this case?
Well, via a scalar quadratic form, the way it should be :)
No, seriously: I didn't have time to read your code carefully. I just
opened the hansl script and that "sqrt" thing jumped at my eye! I may have
some time tomorrow to give it a better look.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
8:14 p.m.
Ah ok, this might be a good hint. I'll check this out.
Artur
Am 10.09.2014 um 21:25 schrieb Riccardo (Jack) Lucchetti:
On Wed, 10 Sep 2014, Artur T. wrote:
> Thanks for the reply, Riccardo. I am aware of this, but I am a bit
> puzzled why it works using gretl's restrict command for this specific
> case I described (restricton three). I am wondered why my function
> yields the same results for the first two restrictions but not for the
> third one. How does gretl exactly compute the Delta method in this case?
Well, via a scalar quadratic form, the way it should be :)
No, seriously: I didn't have time to read your code carefully. I just
opened the hansl script and that "sqrt" thing jumped at my eye! I may
have some time tomorrow to give it a better look.
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
8:27 a.m.
Am 10.09.2014 um 20:54 schrieb Artur T.:
Thanks for the reply, Riccardo. I am aware of this, but I am a bit
puzzled why it works using gretl's restrict command for this specific
case I described (restricton three). I am wondered why my function
yields the same results for the first two restrictions but not for the
third one. How does gretl exactly compute the Delta method in this case?
Artur
I don't yet see any contradiction to what Jack said: For your data
constellation it could be that the negative off-diagonal entries only
happen for the third restriction. So you could have been just lucky that
your approach worked at all, but just in some cases.
I kind of suspect that you're taking the square root "too early".
cheers,
sven
3755
days inactive
3765
days old
5 comments
3 participants
participants (3)
-
Artur T. -
Riccardo (Jack) Lucchetti -
Sven Schreiber