RE: st: RE: Problem with ivreg2 for Panel data

["Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>]

Abubakr,

> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of 
> Abubakr Saeed
> Sent: 04 June 2012 09:11
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: st: RE: Problem with ivreg2 for Panel data
> 
> Hi
> 
> Thank you very much Mark, I got your point and I have 
> thoroughly checked the thread that you suggested. I have one 
> additional question in this regard,
> 
> My endogenous variable is not time-invariate, which is 
> leverage (I am regressing performance against leverage and 
> other variables), In fact, one another important variable on 
> RHS is time-invariate. In this case, even then do I need to 
> follow the same considerations,

In a word, yes.  If you use the FE estimator, you will not be able to obtain an estimate of the coefficient on your time-invariant leverage measure.  You need an estimator that uses between-variation to identify that coefficient.

> or should I estimate with  xtivreg , and run xtoverid for 
> J-stats, but how can I get the Hausman test (that is 
> achievable with ivendog with
> ivreg2 for cross-sectional data) and F-stat.

An estimation strategy you might consider is discussed somewhere in Angrist & Pischke, Mostly Harmless Econometrics (Princeton University Press 2009).  (Sorry, I can't remember exactly where.)  Part of the procedure in assembling the Hausman-Taylor estimator uses this procedure.  It goes something like this:

1.  Use the FE estimator to obtain estimates of beta_fe, the coefficients on the time-varying regressors x_it.  This step will, by the way, also give you a J/overidentification test stat for your endogenous regressors.

2.  Use the estimated beta_fe, the regressors x_it and the dependent variable y_it to calculate

d_it = y_it - x_it'beta_fe

3.  Apply the between estimator to

d_it = f_i'deta_be + u_it

where f_i is your time-invariant regressor (leverage, in your case) and delta_be is the corresponding coefficient.

An advantage of this procedure - depending on your priors about how serious a problem unobserved time-invariant heterogeneity may be - is that you are using the between-variation only where it is absolutely essential, i.e., in the estimation of the coefficient delta_be.

Cheers,
Mark


> 
> Thank you for your useful suggestions.
> 
> Abubakr Saeed,
> 
> 
> On 3 June 2012 22:31, Schaffer, Mark E <M.E.Schaffer@hw.ac.uk> wrote:
> > Abubakr,
> >
> >> -----Original Message-----
> >> From: owner-statalist@hsphsun2.harvard.edu
> >> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Abubakr 
> >> Saeed
> >> Sent: 03 June 2012 11:43
> >> To: statalist@hsphsun2.harvard.edu
> >> Subject: st: Problem with ivreg2 for Panel data
> >>
> >> Hi,
> >>
> >> I am facing trouble in IV estimation. Though in few other 
> threads it 
> >> has already been discussed but it is still not clear to me. 
> >> Basically, I have the Panel data with the time invariate 
> explanatory 
> >> variable.
> >
> > You have the same problem that Benjamin had, discussed here:
> >
> > http://www.stata.com/statalist/archive/2012-05/msg01304.html
> >
> > Your only option is to use an estimator that exploits "between"
> > (cross-sectional)  variation.  Examples are the between 
> estimator and 
> > the random effects estimator.  Also...
> >
> >> I have few questions in this regard:
> >>
> >> 1) Can I use ivreg2 for the Panel data, since Help file says:
> >> "ivreg2 may be used with time-series or panel data, in 
> which case the 
> >> data must be tsset before using ivreg2"-- but running this after 
> >> setting tsset, I am not getting the same results as with xtivreg, 
> >> why?, Is is normal?
> >
> > -ivreg2- on panel data estimates pooled OLS or IV.  -xtivreg- 
> > estimates error components models, e.g., fixed effects, 
> G2LS, EC2SLS etc.
> >
> >> 2) I can not use xtivreg2, since explanatory variable is time 
> >> invariate, xtivreg2 drops it.
> >
> > See above.  That's because -xtivreg2- estimates only fixed 
> effects and 
> > first differences models.
> >
> > --Mark
> >
> >> 3) I prefer to use xtivreg and I have estimated the 
> equation with it 
> >> but how can I get the F-statistics, and Hansen-J statistics while 
> >> using xtivreg.
> >>
> >>
> >> Any response will be much appreciated.
> >>
> >> Regards,
> >>
> >> Abubakr Saeed
> >> PhD student
> >> *
> >> *   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/
> >>
> >
> >
> > --
> > Heriot-Watt University is the Sunday Times Scottish 
> University of the 
> > Year 2011-2012
> >
> > Heriot-Watt University is a Scottish charity registered 
> under charity 
> > number SC000278.
> >
> >
> > *
> > *   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/
> 
> *
> *   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/
> 


-- 
Heriot-Watt University is the Sunday Times
Scottish University of the Year 2011-2012

Heriot-Watt University is a Scottish charity
registered under charity number SC000278.


*
*   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/