st: heterogenous year effects using xtabond2

["Lovisa Persson" <lovisa.persson@nek.uu.se>]

Hello everyone,

I have a question about the xtabond2 command, which is a user written
command.
I am using Stata 12.1 for windows.

I am estimating the consumption sensitivity to income, using a panel data
set on Swedish municipalities.

This is how I write my main command...

xtabond2 consumption income dum1-dum19, gmm(income, laglimits(3 5))
iv(dum1-dum19) robust noleveleq artests(3)

dum1-dum19 is a full set of year dummies. 

Now I want to allow for the coefficient on "income" to be different for
different time periods.  More specifically, if the coefficient on income is
different before and after a reform in year 2000 I am studying.

I can compare estimates before and after 2000 by splitting the sample and
first drop all observations after 1999 and simply run my command, this is
the ?before? estimate.
Then I drop all observations before 1997, and get the ?after? reform
estimate. Why not split at 2000? Well because, the first year ?after? I want
to estimate is year 2000, and for this I need the third lag which is 1997. I
am using years before the reform as instruments, but not estimating any
years before, this is automatic since there are no instruments for these
years.

My question is how to compare these two time periods with an interaction
term, and indicator of which period we are in interacted with income,
instead.

If I construct a dummy for all years from 2000 and onwards, lets call it
?after?, and interact it with income, ?afterinter? I get the following model
to estimate.

xtabond2 consumption income afterinter after dum1-dum19, gmm(income,
laglimits(3 5)) gmm(afterinter, laglimits(3 5)) iv(after dum1-dum19) robust
noleveleq artests(3)

However if ?afterinter? is instrumented only by itself, and not ?income?,
the result will be that three years will be dropped in the estimation of
?afterinter? since it contains only observations from 2000 and onwards and
the first year that can be estimated is 2003. I am then comparing period
1996-1999 with 2003-2011 and not to 2000-2011, which I would want to do.
The estimate for the interaction is indeed not the same as the difference
between the two estimations I described earlier, so the command line is not
equivalent with the sample splitting procedure, as I want it to be.

How can I proceed with the xtabond2 command, allowing for heterogenous
effects in time periods, or even during single years, using an interaction
term.

Thank you for your kind help,

Lovisa Persson
Ph.D. Student
Department of Economics
Uppsala University




*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/