Lukas,
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of
> Lukas Mohler
> Sent: 07 May 2007 14:16
> To: statalist@hsphsun2.harvard.edu
> Subject: st: ivreg2 gmm or cue options and initial values
>
> Dear Statalisters,
>
> I have a question regarding the ivreg2 command. I'd like to
> estimate a model with a constant and two endogenous variables
> which are instrumented by up to 150 instruments (All of them
> are dummies. If you want to know why I do this,
> see: Broda and Weinstein, Globalization and the Gains from
> Trade, The Quarterly Journal of Economics, May 2006). So I
> estimate the following:
>
> ivreg2 y (x1 x2=z1 z2 z3 z4 ... z150), gmm robust
>
> This, of course, is not a problem. But now I want to restrict
> the coefficient of
> x1 to take only positive values (why: see paper mentioned
> above...). In other words I need to perform a grid search
> over values from zero to, say, 150 in specified steps. If I
> understood the help file of ivreg2 correctly, this should be
> possible with the cue option where I can specify the initial values.
I don't think the -cue- option will do what you want with respect to grid searches. It simply minimizes the CUE-GMM objective function using Stata's -ml-. Since -ml- doesn't do a grid seach, the -cue- option won't either.
Stata doesn't readily support inequality constraints. When using -ml- you have to resort to tricks like estimating the log of a coefficient instead of the coefficient itself.
But maybe someone else has ideas here.
> My first question is: Is it correct that the cue option
> allows for a grid search between specified values? If not,
> does anybody know of commands that make that possible or even
> of another software package?
>
> Secondly, if I use the following command:
>
> ivreg2 y (x1 x2=z1 z2 z3 z4 ... z150), cue cueinit(b)
> cueoptions(???) robust
>
> I have read the ml help and tried to pass options to the ml
> routine, but I always get the error message "invalid syntax".
> How can I correctly pass the options to perform my search?
Not sure why you're getting problems here. Here are examples of passing different maximization algorithms using the toy auto dataset:
. sysuse auto
. ivreg2 price (mpg=weight trunk), cue robust cueopt(technique(nr))
initial: neg GMM obj function -J = -.74672932
rescale: neg GMM obj function -J = -.74672932
Iteration 0: neg GMM obj function -J = -.74672932
Iteration 1: neg GMM obj function -J = -.74522321
Iteration 2: neg GMM obj function -J = -.74522317
CUE estimation
--------------
... etc.
. ivreg2 price (mpg=weight trunk), cue robust cueopt(technique(dfp))
initial: neg GMM obj function -J = -.74672932
rescale: neg GMM obj function -J = -.74672932
Iteration 0: neg GMM obj function -J = -.74672932
Iteration 1: neg GMM obj function -J = -.74535713
Iteration 2: neg GMM obj function -J = -.74531459
Iteration 3: neg GMM obj function -J = -.74531458
CUE estimation
--------------
... etc.
-technique(nr)- seems to find a smaller minimum and gets there faster.
--Mark
>
> Thanks.
>
> Lukas
>
>
> ---
> Lukas Mohler
> Abteilung Aussenwirtschaft und Europäische Integration
> Wirtschaftswissenschaftliches Zentrum Universität Basel
> Petersgraben 51
> 4003 Basel
> Switzerland
> +41 61 267 07 70
> ---
>
>
> ----------------------------------------------------------------
> This message was sent using IMP, the Internet Messaging Program.
>
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/