Emanuele,
I heartily concur that you want clustered as opposed to just robust
standard errors for the same reason mentioned already, the existence
of serial correlation in the daily return data. However, note that
you must choose a single dimension for clustering, in this case within
each firm's daily time series of returns and the process assumes
independence across firm returns within a single time dimension, a
trading day, which is not true. From my experience, the daily single
firm return correlation overwhelms the cross-sectional (between firm)
return correlation. You would need to cluster on both dimensions to
eliminate all the dependency. For a great overview on these concepts
take a look at the first paper on Mitchell Petersen's website in the
following link. It helped me greatly:
http://www.kellogg.northwestern.edu/faculty/petersen/htm/working.htm
Tom
On Tue, Jul 1, 2008 at 11:10 AM, Austin Nichols <[email protected]> wrote:
>
> emanuele --
> I would say rather that the reason to use the CRSE is to allow for
> arbitrary serial correlation of errors within panel, and generally I
> would trust the CRSE more than the het-robust SE. Normally, you would
> want 50 clusters or more to ensure that the downward bias of the CRSE
> estimator is negligible, but with 30 balanced clusters, I doubt the
> bias is a problem. So use the CRSE.
>
> On 7/1/08, Rodrigo Alfaro A. <[email protected]> wrote:
> > It is understood to take crse in panels. The reason is simple, you want
> > to consider both the time and cross-sectional variation in the
> > computation of the se's.
>
> > -----Mensaje original-----
> emanuele canegrati
> >
> > Dear all,
> >
> > I am currently writing an econometric paper on the relation between
> > market returns and financial technical indicators (MACD, Relative
> > Strenght Index...). Since the database I am using is a panel of listed
> > companies (around 30 companies; daily observations from January 2003 to
> > March 2008), I decided to use the STATA's option "Clustered Robust
> > Standard Errors". I also run the regressions with normal robust standard
> > errors, obtaining very different results as for the significance of
> > indicators. I wish to ask if you can kindly give me an opinion about
> > which one of the two techniques to use: RSE or CRSE? It seems that by
> > using CRSE I obtain results very similar to those I obtain by performing
> > regressions without robust option.
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
--
Thomas Jacobs
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
