10:26 a.m.
Dear gretl users,
I have a questions regrding coefficient wald tests.
I have estimated a SUR system (20 equations) and now I would like to conduct Wald tests
for individual coefficients across different equations.
I know that I can use the restrict command e.g.
restrict "System"
b[1,2]-b[3,2]=0
end restrict
estimate "System" method=sur
which yields an F-test at the end.
However, I want to conduct the tests for several equations (14 out of 20) each with 8
coefficients, which would require a lot of individual restricitons.
Therefore, I asked myself, whether I could obtain the same results using
(b[1,2]-b[3,2])/S(diff),
where S(diff) =sqrt(S(b[1,2])^2+S(b[1,2])^2)
Unfortunately, the results between the two tests differ (even if I square the
t-statistic).
Finally, I am looking for a way, to calculate the wald tests, without manually
respecifying the every single restriction.
Thanks in advance.
Jan
10:48 a.m.
Am 26.04.2013 12:26, schrieb Jan Tille:
However, I want to conduct the tests for several equations (14 out of
20) each with 8 coefficients, which would require a lot of individual
restricitons.
Therefore, I asked myself, whether I could obtain the same results
using
(b[1,2]-b[3,2])/S(diff),
where S(diff) =sqrt(S(b[1,2])^2+S(b[1,2])^2)
Unfortunately, the results between the two tests differ (even if I
square the t-statistic).
Well in some sense you give the answer yourself, this is something
different. In fact, I don't see what it is or how it is going to work
(in econometric theory I mean), but maybe I'm missing something.
Finally, I am looking for a way, to calculate the wald tests, without
manually respecifying the every single restriction.
Quoting from the command docs, look it up for more explanation:
"
restrict
R = Rmat
q = qvec
end restrict
"
Or maybe you could do it in a loop if you want to somehow test stuff in
different equations.
hth,
sven
11:13 a.m.
In addition to Sven's answer:
You may rewrite restrictions as the one you mention as a t-test, but then your denominator
Sdiff should take into account the covariance between the two coefficients:
sqrt(var(b[1,2])+var(b[3,2])-2cov(b[1,2],b[3,2])).
Good luck,
Jan
Jan Annaert
UNIVERSITEITANTWERPEN | Faculty of Applied Economics (TEW) | Dept. Accounting &
Finance
Room S.B.335 | Prinsstraat 13 | B-2000 Antwerp | Belgium
Phone +32 32654163 |Fax +32 32654064
http://www.ua.ac.be/jan.annaert
http://ssrn.com/author=143473
From: Jan Tille
<Jan.Tille@absolut-research.de<mailto:Jan.Tille@absolut-research.de>>
Reply-To: "gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>"
<gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>>
Date: vrijdag 26 april 2013 12:26
To: "gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>"
<gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>>
Subject: [Gretl-users] Coeffiicent Wald Tests across equations for individual coefficents
(within SUR)
Dear gretl users,
I have a questions regrding coefficient wald tests.
I have estimated a SUR system (20 equations) and now I would like to conduct Wald tests
for individual coefficients across different equations.
I know that I can use the restrict command e.g.
restrict "System"
b[1,2]-b[3,2]=0
end restrict
estimate "System" method=sur
which yields an F-test at the end.
However, I want to conduct the tests for several equations (14 out of 20) each with 8
coefficients, which would require a lot of individual restricitons.
Therefore, I asked myself, whether I could obtain the same results using
(b[1,2]-b[3,2])/S(diff),
where S(diff) =sqrt(S(b[1,2])^2+S(b[1,2])^2)
Unfortunately, the results between the two tests differ (even if I square the
t-statistic).
Finally, I am looking for a way, to calculate the wald tests, without manually
respecifying the every single restriction.
Thanks in advance.
Jan
_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu<mailto:Gretl-users@lists.wfu.edu>
http://lists.wfu.edu/mailman/listinfo/gretl-users
11:58 a.m.
Jan, Thank you for this hint. Is there an easy way to obtain the coefficient cov-matrix,
as Gretel only produces the residiual covariance matrix for the System.
Jan
Am 26.04.2013 um 13:16 schrieb "Annaert Jan"
<jan.annaert@ua.ac.be<mailto:jan.annaert@ua.ac.be>>:
In addition to Sven's answer:
You may rewrite restrictions as the one you mention as a t-test, but then your denominator
Sdiff should take into account the covariance between the two coefficients:
sqrt(var(b[1,2])+var(b[3,2])-2cov(b[1,2],b[3,2])).
Good luck,
Jan
Jan Annaert
UNIVERSITEITANTWERPEN | Faculty of Applied Economics (TEW) | Dept. Accounting &
Finance
Room S.B.335 | Prinsstraat 13 | B-2000 Antwerp | Belgium
Phone +32 32654163 |Fax +32 32654064
http://www.ua.ac.be/jan.annaert
http://ssrn.com/author=143473
From: Jan Tille
<Jan.Tille@absolut-research.de<mailto:Jan.Tille@absolut-research.de>>
Reply-To: "gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>"
<gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>>
Date: vrijdag 26 april 2013 12:26
To: "gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>"
<gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>>
Subject: [Gretl-users] Coeffiicent Wald Tests across equations for individual coefficents
(within SUR)
Dear gretl users,
I have a questions regrding coefficient wald tests.
I have estimated a SUR system (20 equations) and now I would like to conduct Wald tests
for individual coefficients across different equations.
I know that I can use the restrict command e.g.
restrict "System"
b[1,2]-b[3,2]=0
end restrict
estimate "System" method=sur
which yields an F-test at the end.
However, I want to conduct the tests for several equations (14 out of 20) each with 8
coefficients, which would require a lot of individual restricitons.
Therefore, I asked myself, whether I could obtain the same results using
(b[1,2]-b[3,2])/S(diff),
where S(diff) =sqrt(S(b[1,2])^2+S(b[1,2])^2)
Unfortunately, the results between the two tests differ (even if I square the
t-statistic).
Finally, I am looking for a way, to calculate the wald tests, without manually
respecifying the every single restriction.
Thanks in advance.
Jan
_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu<mailto:Gretl-users@lists.wfu.edu>
http://lists.wfu.edu/mailman/listinfo/gretl-users
_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu<mailto:Gretl-users@lists.wfu.edu>
http://lists.wfu.edu/mailman/listinfo/gretl-users
12:13 p.m.
Am 26.04.2013 13:58, schrieb Jan Tille:
Jan, Thank you for this hint. Is there an easy way to obtain the
coefficient cov-matrix, as Gretel only produces the residiual covariance
matrix for the System.
$vcv should do the trick -- while VARs and VECMs are special cases, here
we're talking of a "standard" (=general) system where $vcv should work.
What made you think that only the resid covar was available?
hth,
sven
12:27 p.m.
As far as I remember, there should be a $vcv available.
Jan
From: Jan Tille
<Jan.Tille@absolut-research.de<mailto:Jan.Tille@absolut-research.de>>
Reply-To: "gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>"
<gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>>
Date: vrijdag 26 april 2013 13:58
To: "gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>"
<gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>>
Subject: Re: [Gretl-users] Coeffiicent Wald Tests across equations for individual
coefficents (within SUR)
Jan, Thank you for this hint. Is there an easy way to obtain the coefficient cov-matrix,
as Gretel only produces the residiual covariance matrix for the System.
Jan
Am 26.04.2013 um 13:16 schrieb "Annaert Jan"
<jan.annaert@ua.ac.be<mailto:jan.annaert@ua.ac.be>>:
In addition to Sven's answer:
You may rewrite restrictions as the one you mention as a t-test, but then your denominator
Sdiff should take into account the covariance between the two coefficients:
sqrt(var(b[1,2])+var(b[3,2])-2cov(b[1,2],b[3,2])).
Good luck,
Jan
Jan Annaert
UNIVERSITEITANTWERPEN | Faculty of Applied Economics (TEW) | Dept. Accounting &
Finance
Room S.B.335 | Prinsstraat 13 | B-2000 Antwerp | Belgium
Phone +32 32654163 |Fax +32 32654064
http://www.ua.ac.be/jan.annaert
http://ssrn.com/author=143473
From: Jan Tille
<Jan.Tille@absolut-research.de<mailto:Jan.Tille@absolut-research.de>>
Reply-To: "gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>"
<gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>>
Date: vrijdag 26 april 2013 12:26
To: "gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>"
<gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>>
Subject: [Gretl-users] Coeffiicent Wald Tests across equations for individual coefficents
(within SUR)
Dear gretl users,
I have a questions regrding coefficient wald tests.
I have estimated a SUR system (20 equations) and now I would like to conduct Wald tests
for individual coefficients across different equations.
I know that I can use the restrict command e.g.
restrict "System"
b[1,2]-b[3,2]=0
end restrict
estimate "System" method=sur
which yields an F-test at the end.
However, I want to conduct the tests for several equations (14 out of 20) each with 8
coefficients, which would require a lot of individual restricitons.
Therefore, I asked myself, whether I could obtain the same results using
(b[1,2]-b[3,2])/S(diff),
where S(diff) =sqrt(S(b[1,2])^2+S(b[1,2])^2)
Unfortunately, the results between the two tests differ (even if I square the
t-statistic).
Finally, I am looking for a way, to calculate the wald tests, without manually
respecifying the every single restriction.
Thanks in advance.
Jan
_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu<mailto:Gretl-users@lists.wfu.edu>
http://lists.wfu.edu/mailman/listinfo/gretl-users
_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu<mailto:Gretl-users@lists.wfu.edu>
http://lists.wfu.edu/mailman/listinfo/gretl-users
1:09 p.m.
Sven, Jan
Thank you for this, that was very useful.
Jan
Am 26.04.2013 um 14:30 schrieb "Annaert Jan"
<jan.annaert@ua.ac.be<mailto:jan.annaert@ua.ac.be>>:
As far as I remember, there should be a $vcv available.
Jan
From: Jan Tille
<Jan.Tille@absolut-research.de<mailto:Jan.Tille@absolut-research.de>>
Reply-To: "gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>"
<gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>>
Date: vrijdag 26 april 2013 13:58
To: "gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>"
<gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>>
Subject: Re: [Gretl-users] Coeffiicent Wald Tests across equations for individual
coefficents (within SUR)
Jan, Thank you for this hint. Is there an easy way to obtain the coefficient cov-matrix,
as Gretel only produces the residiual covariance matrix for the System.
Jan
Am 26.04.2013 um 13:16 schrieb "Annaert Jan"
<jan.annaert@ua.ac.be<mailto:jan.annaert@ua.ac.be>>:
In addition to Sven's answer:
You may rewrite restrictions as the one you mention as a t-test, but then your denominator
Sdiff should take into account the covariance between the two coefficients:
sqrt(var(b[1,2])+var(b[3,2])-2cov(b[1,2],b[3,2])).
Good luck,
Jan
Jan Annaert
UNIVERSITEITANTWERPEN | Faculty of Applied Economics (TEW) | Dept. Accounting &
Finance
Room S.B.335 | Prinsstraat 13 | B-2000 Antwerp | Belgium
Phone +32 32654163 |Fax +32 32654064
http://www.ua.ac.be/jan.annaert
http://ssrn.com/author=143473
From: Jan Tille
<Jan.Tille@absolut-research.de<mailto:Jan.Tille@absolut-research.de>>
Reply-To: "gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>"
<gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>>
Date: vrijdag 26 april 2013 12:26
To: "gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>"
<gretl-users@lists.wfu.edu<mailto:gretl-users@lists.wfu.edu>>
Subject: [Gretl-users] Coeffiicent Wald Tests across equations for individual coefficents
(within SUR)
Dear gretl users,
I have a questions regrding coefficient wald tests.
I have estimated a SUR system (20 equations) and now I would like to conduct Wald tests
for individual coefficients across different equations.
I know that I can use the restrict command e.g.
restrict "System"
b[1,2]-b[3,2]=0
end restrict
estimate "System" method=sur
which yields an F-test at the end.
However, I want to conduct the tests for several equations (14 out of 20) each with 8
coefficients, which would require a lot of individual restricitons.
Therefore, I asked myself, whether I could obtain the same results using
(b[1,2]-b[3,2])/S(diff),
where S(diff) =sqrt(S(b[1,2])^2+S(b[1,2])^2)
Unfortunately, the results between the two tests differ (even if I square the
t-statistic).
Finally, I am looking for a way, to calculate the wald tests, without manually
respecifying the every single restriction.
Thanks in advance.
Jan
_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu<mailto:Gretl-users@lists.wfu.edu>
http://lists.wfu.edu/mailman/listinfo/gretl-users
_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu<mailto:Gretl-users@lists.wfu.edu>
http://lists.wfu.edu/mailman/listinfo/gretl-users
_______________________________________________
Gretl-users mailing list
Gretl-users@lists.wfu.edu<mailto:Gretl-users@lists.wfu.edu>
http://lists.wfu.edu/mailman/listinfo/gretl-users
4011
days inactive
4011
days old
6 comments
3 participants
participants (3)
-
Annaert Jan -
Jan Tille -
Sven Schreiber