Hi -
I have a dynamic panel data model with heteroskedasticity, serial correlation
(AR1), and contemporaneous correlation.
I had intended to use one of the Arellano-Bond-Blover type estimators
(xtabond, xtdpd, xtdpdsys) since they were derived to specifically deal with
lagged dependent variables.
Per the Stata manuals ([XT] pg. 25) xtdpd can handle "some autocorrelation in
the idiosyncratic errors." So this seems to be what I need. Yet on [XT] pg. 88
it says, "The moment conditions used by xtdpd are valid only if there is no
serial correlation in the idiosyncratic errors."!!
So my first question would be whether or not I can use xtdpd. If the answer
is no then the rest of my questions are pointless.
I went ahead and used the estimator using my dependent variable in the dgmmiv
() option as per the examples in [XT]. I used my other independent variables
in the iv() option. I received results not too different than that obtained by
using xtpcse.., correlation(psar1) and xtgls.., panels(correlated) corr(psar1).
Thinking that it may be more appropriate to use moment conditions of
additional variables, I added three more continuous variables to dgmmiv(). At
this point the standard errors exploded and nothing was significant.
It seems that xtdpd is VERY sensitive to which variables are used as iv's
and moments. Has any one else observed this phenomenon? Or have I
misunderstood the use of dgmmiv() and iv()? Or perhaps it's a problem with the
autocorrelation in my data?
Finally, a syntax question. If I use L.dv then I lose the first
observation. Since I have all of the lagged values for many years I
constructed my own lagged-dv variable and included that instead. Is there a
problem doing this? Is xtdpd "looking" for a L.dv in the equation
specification?
Your help is greatly appreciated,
Nola
__________________________________
Nola Agha
Department of Sport Management
Isenberg School of Management
121 Presidents Drive, Room 226
University of Massachusetts
Amherst, MA 01003
Ph: 413-545-5637
E-Mail: [email protected]
*
* 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/
