Hello,
I have a panel dataset with 32 panels and 5 time periods.
I am running a Feasible GLS model where I correct for heteroscedasticity
and autocorrelation:
xtgls lnm lnduty, panel(het) corr(ar1).
Because I have only 32 panels, I am inclined to use
a common AR(1) coefficient for all panels in order to preserve degrees of
freedom. However, I am also interested in results that allow for
different autocorrelation coefficients across panels using the option
"corr(psar1)".
When looking at the estimates of rho with the corr(psar1) option, I
observe that the absolute value of
several estimates are actually greater than 1. According to
Kmenta
(1986, p.619) this is not a surprising result when T is small. He
suggests an alternative and consistant estimate of rho(i) whose values are
confined to the interval from -1 to +1 for any sample size.
I have calculated using the residuals from an earlier estimation
that did not correct for autocorrelation. What I would now like to do is
add a constraint in my do file that uses my calculated values of
rho_hat(i) rather than those in the program when the option corr(psar1) is specified.
I have looked at the FAQs on the stata.com website and have had a
fruitless search in the archives. There is one FAQ that approaches my
problem (imposing a single value of rho in a 2-stage Heckman) but I wasn't
able to modify it succesfully for my purposes. I have a feeling I have to
know the formula that is used when the 'psar1' option is engaged but I
don't know where to get that.
Would anyone know if a question like this has been posed before and where I might go to retrieve
it? OR would anyone be able to tell me what to do?
Thanks very much,
Caroline
*
* 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/
