st: Re: Cross sectional dependence panel data random effects

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

I have not checked the code for -xtcsd- recently but my memory is 
that it is not designed for panel data with large N and small 
T.  Implicitly you are testing whether the variance-covariance matrix 
across groups is diagonal.  For your data the matrix is 213 x 213 but 
since the matrix is symmetric it is necessary to compute 22578 
covariances with 10 observations.  This is technically possible using 
pairwise calculation but the result will be a singular matrix and 
this may explain why -xtcsd- refuses to go any further.

You may be better using a panel estimator which is robust to 
cross-section dependence such as the user written procedure -xtscc- 
which implements the Driscoll-Kraay estimator.  However, you should 
bear in mind that allowing for cross sectional dependence in a large 
N, small T dataset will always be difficult unless you are willing to 
make some explicit assumptions about the nature of the dependence 
(i.e. the structure of the covariance matrix).  An illustration is 
spatial panel models which assume a specific pattern of spatial 
(cross sectional) dependence.

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


> ----------------------------------------------------------------------
>
> Date: Fri, 20 Jul 2012 11:41:38 +0200
> From: Rauf Berent <rsberent@gmail.com>
> Subject: st: Cross sectional dependence panel data random effects
>
> Hello,
>
>
> I would really appreciate some help regarding how to check for cross
> sectional dependence in a random effects model when you have a highly
> unbalanced panel.
>
> I have an unbalanced panel (Number of obs: 1560, Groups: 213,
> Observations per group max: 10, avg: 7.3) as my data consists of a lot
> of companies that have been de-listed, merged or newly listed during a
> 10 year period.
>
> I have done a Hausman test and concluded that I should use a random
> effects model.
>
> When checking for cross sectional dependence/contemporaneous
> correlation in a random effects model xttest2 does not seem to work so
> I have downloaded and used a user written program
>
> ssc install xtcsd
>
> The results I get are the following
>
> . xtcsd, pesaran abs
> Error: The panel is highly unbalanced.
> Not enough common observations across panel to perform Pesaran's test.
> insufficient observations
>
> . xtcsd, friedman
> no observations
>
> xtcsd, frees
> no observations
>
> I am not really sure how to proceed as I was under the impression that
> an unbalanced panel was not a problem for Pesaran. I also was under
> the impression that friedman and frees tackle unbalanced panels by
> only using observations available for all cross sectional units. Being
> new to Stata and econometrics I would really appreciate all the help I
> can get.
>
> Thanks in advance,
>
> Rauf

*
*   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/