Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: overid after reg3


From   "Demetris Christodoulou" <D.Christodoulou@econ.usyd.edu.au>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: overid after reg3
Date   Wed, 2 Jul 2008 22:25:39 +1000

Hello again,  I have just realised that my posting a few days ago overllapped with the German SUG meeting and may have gone unoticed.

Just to make sure, I thought to give it another go in case someone can help out with the issue described below:

many thanks, Demetris

________________________________

From	   "Demetris Christodoulou" <D.Christodoulou@econ.usyd.edu.au <mailto:D.Christodoulou@econ.usyd.edu.au> >	 
To	   <statalist@hsphsun2.harvard.edu <mailto:statalist@hsphsun2.harvard.edu> >	 
Subject	   st: overid after reg3	 
Date	   Mon, 30 Jun 2008 09:10:01 +1000	

________________________________

Hi there, 
 
I estimate the following structural equation and test for overidentifying restrictions:
 
cons 1 ([x1]L1.x1 + [x1]L1.x3 + [x1]L1.x4 + [x2]L1.x2 + [x2]L1.x5 + [x2]L1.x6) = 0
reg3 (y x1 L1.x1 x2) (x1 L1.x1 L1.x3 L1.x4 ind*) (x2 L1.x2 L1.x5 L1.x6 ind*), small dfk l(95) cons(1)
overid
 
but, disappointingly enough, I get the message:
 
L.SAL_d invalid name
               st_view():  3500  invalid Stata variable name
            overidreg3():     -  function returned error
                 <istmt>:     -  function returned error
r(3500);


I make an educate guess and generate lagged values instead of using lagged operators and rerun the exercise, as follows:
 
cons 1 ([x1]Lx1 + [x1]Lx3 + [x1]Lx4 + [x2]Lx2 + [x2]Lx5 + [x2]Lx6) = 0
reg3 (y x1 Lx1 x2) (x1 Lx1 Lx3 Lx4 ind*) (x2 Lx2 Lx5 Lx6 ind*), small dfk l(95) cons(1)
overid
 
only to get another message:
 
                       *:  3900  unable to allocate real <tmp>[14356,14356]
            overidreg3():     -  function returned error
                 <istmt>:     -  function returned error
r(3900);
 
I am in the dark here and any suggestions why this is happening are most welcome. 
Also, if this problem is indeed due to an OS limitation and cannot be circumvented, then is it safe to run ivreg2 regressions and test overid separately? I went through previous Statalist posts and have read that it is possible. Specifically, Mark Schaffer noted "you don't end up in exactly the same place, but it's close" (http://www.stata.com/statalist/archive/2004-09/msg00234.html <http://www.stata.com/statalist/archive/2004-09/msg00234.html> ).
Is "close" good enough? Doesn't this depend on whether the structural system is estimated using 2SLS or 3SLS? 
 
many thanks, Demetris Christodoulou





© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index