Re: st: fixed effect correcting auto correlation and heteroskedasticity

[Gordon Hughes <G.A.Hughes@ed.ac.uk>]

The routine -xtscc- is an implementation of the Driscoll-Kraay robust 
covariance estimator, but it primarily intended to be robust with 
respect to spatial or similar correlation between cross-section 
units.  It is basically a Newey-West type estimator with respect to 
time.  You should read Hoechle's article on -xtscc- in Stata Journal 
referred to in the help information for the program.  You can also 
refer to the original Driscoll & Kraay article in Review of Economics 
& Statistics 1998.  A couple of points:

1.  From your description  I don't think that you have enough time 
periods for any adjustment for serial correlation to be relied 
upon.  However, this is a little complicated because of the way in 
which -xtscc- operates.  In effect it applies a Newey-West estimator 
to a time sequence of cross-section averages, so it depends upon 
precisely how unbalanced your panel is, but if you have a maximum of 
12 periods (for T) you really aren't going to get very much value out 
of using it.

2.  The default specification for -xtscc- is random effects.  There 
is a fixed effects option - again, read the help file.

3.  Since the real point of -xtscc- is cross-sectional spatial 
correlation, which does not appear relevant, you could consider 
-newey2- which is another user-written procedure which implements the 
Newey-West estimator for panel data (among other things).

4.  Given the fact that you have (relatively) large N and small T, 
why not bootstrap the estimation of the covariance matrix (using 
panel units as the population to be re-sampled)?

Gordon Hughes
g.a.hughes@ed.ac.uk

*
*   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/