Re: st: re: xtivreg

[Austin Nichols <austinnichols@gmail.com>]

W Zang :
As an example of Kit's point, consider:

clear all
u http://fmwww.bc.edu/ec-p/data/macro/abdata
tsset id year
xtivreg2 ys k (n=l2.n l3.n), fe endog(n)
qui ivreg2 ys k (n=l2.n l3.n)
ivendog n
qui xi:ivreg2 ys k (n=l2.n l3.n) i.id
ivendog n
testparm _I*

where the last line demonstrates the need for FE, and the endog test
does not support the use of the exclusion restriction--but that does
not necessarily mean that n is exogenous; with a better exclusion
restriction, the endog test could produce a different result.
Remember that the endog test
xtivreg2 ys k (n=l2.n l3.n), fe endog(n)
just compares
xtivreg2 ys k (n=l2.n l3.n), fe
to
xtivreg2 ys k n (=l2.n l3.n), fe
not to another specification with better instruments.


On Wed, May 5, 2010 at 10:07 AM, Christopher Baum <baum@bc.edu> wrote:
> <>
> If the XT-IV model is properly specified, and fixed effects are necessary to ensure consistency, the pooled IV results are inconsistent, and any test based upon them is invalid.  You should be able to use the same syntax with endog(y2) on the ivreg2 command, but I am not surprised that the results differ. In the presence of unobserved heterogeneity, the pooled IV will be junk.
>
> Kit
>
> Kit Baum   |   Boston College Economics and DIW Berlin   |   http://ideas.repec.org/e/pba1.html
> An Introduction to Stata Programming   |   http://www.stata-press.com/books/isp.html
> An Introduction to Modern Econometrics Using Stata   |   http://www.stata-press.com/books/imeus.html
>
> On May 5, 2010, at 9:37 AM, W Zang wrote:
>
>> Dear Kit
>>
>> Thank you very much for your previous reply. I have another question and wonder if you could help me please. Before estimating the two simultaneous equations, I did the endogeneity test of y2.
>> xtivreg2 y1 (y2 = x4 x5) x1 x2 x3, fe endog (y2)
>>
>> I also performed Durbin-Wu-Hausman test.
>> ivreg2 y1 (y2 = x4 x5) x1 x2 x3
>> ivendog (y2)
>>
>> I got two different results. y2 is proved to be exogeneous in the first case and y2 is endogenous in the second case. Did I do something wrong? Which test should I follow?
>>
>> Best Wishes,
>>
>> Linda
>>
>>>
>>> Quoting Kit Baum <baum@bc.edu>:
>>>
>>>> <>
>>>> Linda said
>>>>
>>>> I am trying to use xtivreg to estimate two-stage least squares
>>>> simultaneous equations. I have two endogenous variables Y1 and Y2. Y1
>>>> and Y2 are affected by different exogenous variables. Y1 is affected
>>>> by X1, X2, X3 and Y2 is affected by X1, X4, X5.
>>>>
>>>> Y1 = a0 + a1*Y2 + a2*X1 + a3*X2 + a4*X3
>>>> Y2 = b0 + b1*Y1 + b2*X1 + b3*X4 + b4*X5
>>>>
>>>>
>>>> xtivreg y1 (y2 = x4 x5) x1 x2 x3, fe
>>>> xtivreg y2 (y1 = x2 x3) x1 x4 x5, fe
>>>>
>>>> The syntax is the same as it would be for a non-panel IV estimator. Just list the variables excluded from each equation within the parenthesized expression.

*
*   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/