Re: Re: st:Normality and Granger Causality in Panel data

[Christopher Baum <kit.baum@bc.edu>]

<>
On Apr 3, 2012, at 2:33 AM, DD wrote:

> ell, my panel stracture is: T=10, N=9 (i.e. short panel).
> 
> The reason for asking about a valid Normality test is to be able to 
> reliably use t-statostics on coefficients etc.
> 
> Regarding your question about cross secctional dependence. This can be 
> tested, among other tests, by Pasaran CD test or Breusch-Pagan LM test. I 
> am not sure how you tackle cross sectional dependence. However, Hoechle 
> suggests to use Drisdoll and Kraay standard errors using the command 
> 'xtscc'.
> 
> I am still in search for a valid Normality test for estimating xi: 
> xtreg........., fe/re vce models.

There is no need to agonize over the normality of residuals from this model. Just estimate the model
without the assumption of i.i.d. errors. In xtreg, fe, you can specify the vce() option, and employ
robust, HAC, cluster, bootstrap, or jackknife SEs.

I would not recommend applying the cluster-robust estimator here with a very small T and N. You
could conceptually apply two-way clustering (supported in -xtivreg2- from SSC) if you had larger
T and N, but you don't.

As for Granger causality, it's not clear how to think about that in a panel context. One variable
could G-cause another in one panel and not in another. This is essentially the same problem as
a panel unit root test: how do you compute a box score from individual panels' results?

Kit

Kit Baum   |   Boston College Economics & DIW Berlin   |   http://ideas.repec.org/e/pba1.html
                             An Introduction to Stata Programming  |   http://www.stata-press.com/books/isp.html
  An Introduction to Modern Econometrics Using Stata  |   http://www.stata-press.com/books/imeus.html


*
*   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/