Dear all,
I'm trying to estimate a system of equations using a SURE methodology.
My work is based on the idea that my two dependent variables share
similar characteristics and are therefore affected by a similar set of
independent variables. I assume that the error terms of the two eq.
are somehow linked (e.g. unobsevables may affect both equations) and
this is confirmed by the results of a Breusch-Pagan test for
independent equations.
I am working with a panel setting of 40 countries and 9 years for both
the equations.
I have encountered two kinds of problem:
1. In both equations I have detected groupwise heteroskedasticity (via
the xttest3 command). Going into the statalist archive, I have found
that the command mysureg allows to correct standard error for
intragroup correlation adopting maximum likelihood.
I have also tried with the command _robust on predicted standard
errors, but this gives back different results (too perfect in my
opinion!). Is my method correct?
predict st1, equation(#1) stdp
predict st2, equation(#2) stdp
_robust st1 st2, cluster(panel) minus(50)
ereturn display
2. In order to take into account for panel fixed effects, I have
adopted the within transformation, but as far as I know this requires
to adjust the var-cov matrix for the degrees of freedom lost due to
the exclusion of fixed effects.
In this case I have no idea how to programme this in stata with two equations.
Please let me know if you need any additional information.
Any suggestion is more than welcome, many thanks in advance.
best,
Marco
*
* 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/
