Following email conventions, the response precedes the Q:
Note that -ivreg2- and -overid- and -ivendog- are all from the
same source, the development team of C. F. Baum, M. E.
Schaffer, and S. Stillman, and appear in SJ 3(1):1--31 and
SJ 4(2):224 and SJ 5(4):607 (hence are available from SSC).
The main advantage of -ivreg2- is that it is just like -ivreg- but
with bells and whistles for IV, and more models built in.
What I am recommending is a test for "weak instruments"
which is entirely distinct from testing for endogeneity or using
the overidentification restrictions to test the validity of subsets
of excluded instruments. You should perform *all three* types
of tests whenever you run an instrumental variables model.
When you specify the option ffirst, the minimum eigenvalue of
the Cragg-Donald statistic is reported, and you can compare
that number to the table in Stock and Yogo (2002) to see how
much you have to worry about weak instruments. This start
of this explanation is in the help file for -ivreg2- and the middle
is in Stock and Yogo (2002):
Stock, J. H. and M. Yogo. 2002.
"Testing for weak instruments in linear IV regression."
NBER Technical Working Paper 284.
http://www.nber.org/papers/T0284.
The end is still being written:
http://ksghome.harvard.edu/~JStock/ams/websupp/index.htm
On 4/6/06, Philip Rego <[email protected]> wrote:
> Thanks a lot for your suggestion. I have never used ivreg2 and i do
> not know how it is different from ivreg. even after ivreg, i am
> testing for endogeneity and overidentification like ivendog and
> overid.
>
> so is it require to do anymore test.
>
> thanks in advance.
>
> best,
> philip
>
> On 4/6/06, Austin Nichols <[email protected]> wrote:
> > That is correct, I suppose, in the sense that every excluded
> > instrument is used to form the linear projection, but you will
> > certainly want to use ivreg2 (type -ssc install ivreg2- to get it) and
> > check that you do not have "weak instruments." Even if you pass the
> > Stock & Yogo test, I would be suspicious that your equation is
> > underidentified, and you might also have a hard time convincing anyone
> > that the lag of Z and powers thereof do not affect Y (except through
> > the effect on current Z) but the current values do. Depends on your
> > model, I suppose.
> >
> > On 4/5/06, Philip Rego <[email protected]> wrote:
> > > Hi,
> > > I want to estimate a model of this sort.
> > > Y=constant+c1(z)+c2(z-square)+c3(z-cubed)+error
> > > I want to use IV-reg approach where lag of Z (i.e lag of Z, lag of
> > > z-square, lag of z-cube) are the exogenous variable. My question:
> > > in stata i am writing command like this:
> > >
> > > ivreg y independent variable (z z-square z-cubed=lag of Z, lag of
> > > z-square, lag of z-cube)
> > >
> > > Is it right?
> > > In this way, even z-square's predictors are lag of Z, lag of z-square,
> > > lag of z-cube and so also for z-cubed.
> > >
> > > is it ok.
> > > best,
> > > philip
*
* 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/
