st: Cross-sectional dependence in fixed effects panel model

[christina sakali <christina.sakali@googlemail.com>]

Dear Statalisters,

I would like to estimate a fixed effects panel model with possible
cross-sectional dependence (contemporaneous correlation). My sample
consists of N=11, T=11.

My question regards both the interpretation of the tests contained in
xtcsd and the estimation with Driscoll-Kraay S.E. under xtscc:

1. First, I am not completely sure how I should interpret results of
the tests contained in xtcsd and xttest2, given the fact that I got
some conflicting results from these tests, which are provided below.
My feeling is that I should trust the results of the Pesaran test (as
more appropriate for my sample), which indicates the presence of
cross-sectional dependence, but again I am a little confused regarding
the conflicting and inconclusive results from the other tests.

2. Second, if my model does suffer from cross-sectional dependence, I
am wondering whether the Driscoll-Kraay S.E. (xtscc) are appropriate
for my case. De Hoyos and Sarafidis (2006, p.1-2) clearly mention that
if "cross-sectional dependence is caused by the presence of common
factors, which are unobserved ... but uncorrelated with the included
regressors, ...  one may chose to rely on standard FE/RE methods and
correct the SE by following the approach proposed by Driskoll and
Kraay (1998). On the other hand, if the unobserved components that
create interdependencies across cross-sections are correlated with the
included regressors, these approaches will not work and the FE and RE
estimators will be biased and inconsistent. In this case, one may
follow the approach proposed by Pesaran (2006).

Based on the above, I would like to ask if there is a way to find out
which of the two cases mentioned by De Hoyos and Sarafidis (2006) is
relevent for my model and what other alternatives I 've got in Stata,
in order to produce robust S.E. in presence of contemporaneous
correlation, apart from the Driskoll and Kraay S.E.

In case this is relevant to the above, the residuals in my model seem
to be heteroscedastic but not serially correlated (after checking the
stats in xttest 3 and xtregar ..., fe lbi)

I am using Stata 9.2 at home but I could possibly have access to some
newer version of Stata at faculty.


Results from xtcsd and xttest2 are provided below:

qui xtreg fdi gg rulc trade tert sec eu, fe

. xtcsd, pesaran abs

Pesaran's test of cross sectional independence =     2.899, Pr = 0.0037

Average absolute value of the off-diagonal elements =     0.302

. xtcsd, friedman

 Friedman's test of cross sectional independence =    17.537, Pr = 0.0633

. xtcsd, frees

 Frees' test of cross sectional independence =     0.244
|--------------------------------------------------------|
  Critical values from Frees' Q distribution
                      alpha = 0.10 :   0.2333
                      alpha = 0.05 :   0.3103
                      alpha = 0.01 :   0.4649

. xttest2

Breusch-Pagan LM test of independence: chi2(55) =    79.711, Pr = 0.0164
Based on 11 complete observations over panel units
*
*   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/