[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: xtabond
Date sent: Tue, 16 Sep 2003 12:08:31 -0400 (EDT)
From: Mine Zeynep Senses <firstname.lastname@example.org>
Subject: Re: st: xtabond
Send reply to: email@example.com
> Thats how I did it initially but I don't think its right to do it that
> way. Because year dummies are strictly exogenous and there is no need to
> instrument for them.
I'm not sure about this. If you interact year and cash_k, and you
believe that year is exogenous and cash_k is endogenous, then
yearcash_k will be endogenous, no?
To take a simpler example, say you have a model where you have one
regressor, x, that you believe to be endogenous. You think that two
groups of the population might have different coefficients for this
endogenous variable, so you interact x with a dummy for the
charactertistic D defining the two groups, and you get two
coefficients, one on x and the other on D_x. Both x and D_x are
endogenous and need instrumenting, because if x is not orthogonal to
the error term, then unless you've got good reason to believe
otherwise, D_x will also not be orthogonal to the error term.
> Plus lag of the interaction variable will always be
> zero because of the year dummy, so I am not even sure if it is a valid
I think this is the key point. What is the appropriate set of lags
for the interaction of year and cash_k?
My first reaction would be to say lag(1) of year * lag(1) of cash_k =
lag(1) of cash_k, lag(2) of year * lag(2) of cash_k = lag(2) of
cash_k, and so forth. That is, if the year dummy = 1 this year, then
the lagged dummy = 1 last year. This implies that you can't use lags
of interacted variable yearcash_k for instruments, you use lags of
the uninteracted cash_k instead.
Does this make sense?
> In the literature people usually interact cash variable with
> a size dummy or something, which is time invariant, and then there is no
> problem doing it that way...
> But if you think I am not right on this, Iwill be happy to hear why since
> I got very good results doing it that way and I will be happy to keep
> those results!!!
> > > > I have a question regarding the xtabond command. I am using:
> > > > xi: xtabond investment_2, lags(1) maxlags(3) pre(sales_k, lag(0,3))
> > > > pre(cash_k, lag(0,3)) diffvars(i.year i.industry)
> > > >
> > > > I want to add an interaction variable: year*cash_k. Year is a dummy
> > > > variable and cash_k is predetermined. So what i would like is to
> > > > instrument this variable with year*cash_k(lag) etc. Defining it
> > > > predetermined will use (year*cash_k)(lag) which is probably not right.
> > Can you explain why you think this isn't right? Put another way, I
> > don't see why you can't define a new variable called yearcash_k which
> > is the interaction of year and cash_k and simply work with that.
> > --Mark
> > > >
> > > > is there a way to do this? I will appreciate any help.
> > > > mine
> > >
> > > *
> > > * 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/
> > Prof. Mark E. Schaffer
> > Director
> > Centre for Economic Reform and Transformation
> > Department of Economics
> > School of Management & Languages
> > Heriot-Watt University, Edinburgh EH14 4AS UK
> > 44-131-451-3494 direct
> > 44-131-451-3008 fax
> > 44-131-451-3485 CERT administrator
> > http://www.som.hw.ac.uk/cert
> > *
> > * 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/
Prof. Mark E. Schaffer
Centre for Economic Reform and Transformation
Department of Economics
School of Management & Languages
Heriot-Watt University, Edinburgh EH14 4AS UK
44-131-451-3485 CERT administrator
* For searches and help try: