Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: AW: RE: RE: nlcom from several estimations


From   "Martin Weiss" <martin.weiss1@gmx.de>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: AW: RE: RE: nlcom from several estimations
Date   Mon, 8 Feb 2010 17:34:57 +0100

<> 

" The only way to program N3SLS in Stata would have been to program GMM with
Mata."



Stata 11 comes with a -gmm- command (http://www.stata.com/stata11/gmm.html)
that let`s you do this much more conveniently...


HTH
Martin


-----Ursprüngliche Nachricht-----
Von: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] Im Auftrag von Valerie Orozco
Gesendet: Montag, 8. Februar 2010 16:58
An: statalist@hsphsun2.harvard.edu
Betreff: st: RE: RE: nlcom from several estimations

Martin, maybe I'm wrong, but as far as I know : 
-"reg3" and "nlsur" don't estimate N3SLS with instrumental variables I
wanted to estimate.
-"reg3" allows IV but doesn't treat nonlinearity in parameters.
-"nlsur" allows nonlinearity but no IV. 
The only way to program N3SLS in Stata would have been to program GMM with
Mata. 
But I decided to run the estimation with another software estimating N3SLS
quite easily.
So I'm not able to estimate it with Stata and the only thing I can do is
trying to reconstruct the estimation by "hand".

Yes, I try the "ereturn post" command, even with a small program (Post)
trying to add some local... see below : 


clear
mat drop _all

capture program drop Post
program Post, eclass
        syntax, r(string) vce(string)

        eret post `r' `vce'         
        eret local cmd "nlcom"      
end

/****example with 2 estimations****/

/*1st estimation*/
matload b1		/*in reality, I have a .txt that I insheet, then
mkmat and then matsave*/
matload V1
Post, r(b1) vce(V1)
est store toto

/*2nd estimation*/
matload b2
matload V2
Post, r(b2) vce(V2)
est store titi

suest toto titi 

after "suest" I have the message : "estimation sample of the model saved
under toto could not be restored"
Probably because I haven't the "e(sample)" information I'm not able to have
since I have no data here...only matrix in my Stata session...



-----Message d'origine-----
>From 	  "Martin Weiss" <martin.weiss1@gmx.de>
To 	  <statalist@hsphsun2.harvard.edu>
Subject 	  st: AW: RE: nlcom from several estimations
Date 	  Mon, 8 Feb 2010 16:36:12 +0100

<> 



Have you actually tried to -ereturn post- foreign results and apply -suest-
to them? That sounds like a stretch to me. You should first check whether
you can recreate the results in Stata, perhaps using -reg3/nlsur-?



HTH
-----Message d'origine-----
De : Valerie Orozco 
Envoyé : lundi 8 février 2010 16:29
À : 'statalist@hsphsun2.harvard.edu'
Objet : RE: nlcom from several estimations

Thank you John and Martin.

John :  by "model", I don't mean different groups, but really different
models. I'm estimating 4 different systems of equations (N3SLS= Nonlinear 3
stages system of equations) and the variables are not the same in each one.
(my goal is then to estimate unconditional elasticities in a demand system
coming from the elasticities from different steps...).


"Suest" seems to do what I need except that I have a difficulty due to the
fact that the estimation wasn't done with Stata (because Stata doesn't
easily estimate N3SLS). But like I prefer Stata, I stored the b and V matrix
into text files and then insheet in Stata and re"build" my matrix b`i' and
V`i' (i=1,...,4)...I can also use "ereturn post" to "re"build each
estimation and store it.

But, doing "suest", it seems some informations (for example e(sample),
e(N)...stored by local during the estimation) are missing...

If you have more ideas...

Thank you
Valérie

____________________________________________________


>From 	  "Martin Weiss" <martin.weiss1@gmx.de>
To 	  <statalist@hsphsun2.harvard.edu>
Subject 	  st: AW: nlcom from several estimations
Date 	  Mon, 8 Feb 2010 14:35:14 +0100


Have you had a look at -suest- yet?

HTH
Martin

____________________________________________________

On 08.02.2010 14:32, Valerie Orozco wrote:

    Hello,


I have to estimate some nonlinear combinations of parameters estimated from
several models.
More precisely, I estimated 4 models (4 systems of equations) and have the
corresponding matrix of estimated parameters (b1, b2, b3 and b4) and each
corresponding variance covariance matrix (V1, V2, V3, V4). I want to
estimate combinations of these parameters coming from these 4 estimations.
The problem is that, with "nlcom", it has to be from only one estimation.


So, my first idea was to build two new matrix (b and V) making Stata
believes it comes from only one estimation : 

	- b : just appending b1,b2, b3 and b4 horizontally
    	- V : a block diagonal matrix (each block being V1, V2, V3 and V4)
    	- but of course, some elements of this V block diagonal matrix are
missing

Another idea I had was to use "eststo" and "estout" commands but it seems
the matrix stored correspond to the last estimation.

Do you have some better ideas to help me?

Many thanks.

Valérie



-------------------------------
Valérie OROZCO
Toulouse School of Economics (INRA-GREMAQ)
21, allée de Brienne
F-31000 Toulouse, France

MA 101
+33 5 61 12 85 91
-------------------------------


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



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