|
|  |
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: RE: RE: hausman's test with xtivreg2
Erasmo,
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of
> Erasmo Giambona
> Sent: Friday, April 20, 2007 10:54 AM
> To: [email protected]
> Subject: st: RE: hausman's test with xtivreg2
>
> Dear Mark,
> this was really helpful. Regarding my last question, i was
> referring to the Hausman test for endogeneity.
> There is a paper by Himmelberg et al. (Journal of Financial
> Economics (one of the leading finance journals), 53 (1999)
> pp: 353-384) which on page 373 states that the Hausman's test
> tends to overly reject the null hypothesis of exogeneity if
> any of the explanatory variables are correlated with the
> fixed effects. That is, the test would be affected by Type I
> error. A copy and paste of the paragraph from the paper follows:
>
> "One can formalize this evidence against the exogeneity of
> managerial ownership by testing for a correlation between the
> fixed effect and managerial
> ownership. We could use a Hausman (1978) test, but this test
> would tend to
> over-reject the null hypothesis of zero correlation because
> it would tend to reject
> if any of the explanatory variables were correlated with the
> fixed effect.
> To reduce this Type I error, we construct a more precise
> &conditional moment'
> test, which is in the spirit of a Hausman test, but tends to
> reject only if
> managerial ownership is the source of the specification error
> (Greene, 1997, p. 534; Newey, 1985)."
This is very curious. Do we have the context right?
If I'm not mistaken, the above makes sense if the authors are talking
about the application of a Hausman test to an equation without fixed
effects. In that case, omission of the fixed effects would generate
omitted variable bias if the explanatory variables are correlated with
the omitted "fixed effects" since they are now in the error term.
But say we apply the Hausman test to a fixed effects estimation done as
a LSDV (least squares dummy variable) estimation. Then I can't see what
the problem would be. There's nothing wrong with explanatory variables
being correlated with each other, and the LSDVs that are included as
regressors are just that.
--Mark
>
>
> Thanks very much,
>
> Erasmo
> *
> * 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/
