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,JanJan 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 32654064http://www.ua.ac.be/jan.annaert
http://ssrn.com/author=143473
From: Jan Tille <Jan.Tille@absolut-research.de>
Reply-To: "gretl-users@lists.wfu.edu" <gretl-users@lists.wfu.edu>
Date: vrijdag 26 april 2013 12:26
To: "gretl-users@lists.wfu.edu" <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]=0end 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 mailing list
Gretl-users@lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users