Re: st: GMM error (bug in Stata?)

[John Antonakis <John.Antonakis@unil.ch>]

Hi Stas:

Thanks for your note.

Yes; the main equations identical but that has not stopped gmm before; 
run the following:
clear
set seed 123
set obs 1000
gen x = rnormal()
gen z = rnormal()
gen q = rnormal()
gen y1 = x + z + q + rnormal()
gen y2 = y1 + q + rnormal()

gmm (eq1: y2 - {b1}*y1 - {b0}) ///
    (eq2: y2 - {c1}*y1 - {c0}), ///
    instruments(eq1: x) ///
    instruments(eq2: z) ///
    twostep winitial(unadjusted, indep)
    test [b1]_cons = [c1]_cons

The goal of my estimation procedure is to make cross model comparison, 
where the model have different instruments ( and given the clustering I 
have, I want to have a generalized Hausman test hence the use of gmm). I 
want to show that the second stage estimates don't change when I change 
the instruments....it's a simulation study I am working on, hence the 
"strangeness".
So, it seems that I don't have an error in my initial code then?

Best,
J.

__________________________________________

Prof. John Antonakis
Faculty of Business and Economics
Department of Organizational Behavior
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny
Switzerland
Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305
http://www.hec.unil.ch/people/jantonakis

Associate Editor
The Leadership Quarterly
__________________________________________


On 29.10.2011 18:42, Stas Kolenikov wrote:
> On Sat, Oct 29, 2011 at 11:33 AM, John Antonakis 
<John.Antonakis@unil.ch> wrote:
>> Hi:
>>
>> I am trying to estimate "stacked" models using -gmm-. When I 
estimate the
>> two models separately, things work fine (see code on the bottom of my
>> e-mail). However, when I run the models jointly, with the following
>> code...........
>>
>> gmm  ///
>> (eq1: y - {b1}*x_style1 - {b2}*x_style2- {b3}*x_style3- 
{b4}*x_style4- ///
>> {b5}*x_style5- {b6}*x_style6 - {b7}*x_style7- {b8}*x_style8- 
{b9}*x_style9-
>> ///
>> {b10}*x_style10- {b11}*x_style11- {b12}*x_style12- {b13}*x_style13- 
{b0})
>> ///
>>
>> (eq2: y - {c1}*x_style1 - {c2}*x_style2- {c3}*x_style3- 
{c4}*x_style4- ///
>> {c5}*x_style5- {c6}*x_style6 - {c7}*x_style7- {c8}*x_style8- 
{c9}*x_style9-
>> ///
>> {c10}*x_style10- {c11}*x_style11- {c12}*x_style12- {c13}*x_style13- 
{c0}),
>> ///
>>
>> instruments(eq1: x_fe1 x_fe2 x_fe3 x_fe4 x_fe5 x_fe6 x_fe7 x_fe8 
x_fe9 ///
>> x_fe10 x_fe11 x_fe12 x_fe13 ) ///
>>
>> instruments(eq2:  x_clus1 x_clus2 x_clus3 x_clus4 x_clus5 x_clus6 
x_clus7
>> ///
>> x_clus8 x_clus9 x_clus10 x_clus11 x_clus12 x_clus13) ///
>>
>> twostep winitial(unadjusted, indep) vce(cluster lead_n)
>>
> John, are the main equations identical? That's a strange model to
> estimate, then. At the very least, the orthogonality of residuals from
> the first equation and from the second equations to the intercept
> would imply identical moment conditions, and that would be enough for
> GMM to break down.
>

*
*   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/