Hi Mark,
Thanks a lot for your prompt reply. Your comments are very helpful.
Sandra
***********************************************
Sandra Mortal
Assistant Professor of Finance
516 Cornell Hall
University of Missouri
Columbia, MO 65211
Office: (573) 884-1684
Fax: (573) 884-6296
***********************************************
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Schaffer,
Mark E
Sent: Wednesday, January 24, 2007 7:15 PM
To: [email protected]
Subject: st: RE: FW: Instrumental variables using two dummy variables as
instruments
Sandra,
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of Mortal,
> Sandra C.
> Sent: 25 January 2007 00:40
> To: [email protected]
> Subject: st: FW: Instrumental variables using two dummy variables as
> instruments
>
> Hi,
>
> I am running a regular OLS regression, but need to use instrumental
> variables because of an endogenous variable.
> This variable is continuous, but I am only able to think of two dummy
> variables to use as instruments (it is very difficult to find
> instruments for this variable). These seem to be good instruments
> because they are theoretically exogenous, and they have a strong
> correlation with the endogenous variable (F tests that these variables
> are equal to zero are well above 10). The problem is that in the
> second stage regression the coefficient of the instrumented variable
> is much higher than that of the endogenous variable in the plain OLS
> regression. I wonder if you could point me in the right direction.
> Specifically, are there problems with using dummy variables as
> instruments,
No, there aren't, or at least not in your context and nothing that you
haven't already thought about, e.g., the correlation with the endogenous
regressor.
> and do you know of
> a reason why the second stage regression coefficients are so high?
This may be perfectly reasonable, or it may be generated by invalid
instruments.
The sign of the OLS bias may be + or -, depending on the sign of the
correlation of the endogenous variable with the error term. If you are
using invalid instruments, they too can be biased, and again the
direction can be + or -. Thus, you can have b_IV > b_OLS because the
latter is biased downwards, the former is biased upwards, or both.
Wooldridge's u/g econometrics text, Introductory Econometrics, has a
clear discussion of this, though I don't have the page references handy.
By the way, since you have an overidentified equation, you should check
to see if it passes an overidentifying restrictions test.
Also by the way, you may want to use -ivreg2- for your estimation, since
it generates an overid stat automatically (as well as some other useful
diagnostics). Or you can use -overid- to get the overid stat.
> Is this normal?
> I am sorry this is not a directly STATA related question,
Last BTW - it's "Stata", not "STATA".
Cheers,
Mark
> but I have exhausted all of my other options, given my limited
> knowledge of statistics.
>
> Thank you,
> Sandra
>
>
>
>
> ***********************************************
> Sandra Mortal
> Assistant Professor of Finance
> 516 Cornell Hall
> University of Missouri
> Columbia, MO 65211
> Office: (573) 884-1684
> Fax: (573) 884-6296
> ***********************************************
>
>
>
>
> *
> * 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/
*
* 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/
