RE: st: RE: Cluster standard errors by time and firm

["Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>]

Sebastiaan,

> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of 
> sebas nicaise
> Sent: 11 June 2012 17:39
> To: statalist@hsphsun2.harvard.edu
> Subject: RE: st: RE: Cluster standard errors by time and firm
> 
> Again, not very well pointed out by me. I meant that I have a 
> maximum of 40 quarters per firm but since it is unbalanced I 
> have aproximately 180000 observations.

In this context, the total number of observations is less relevant than
the numbers in the N and T dimensions.

>  So I could do 2-way clustering in stata?

2-way clustering relies on both N and T being large.  Your N (7k) is
large enough, but your T ("max" 40) is borderline.

Can you say more about the number of observations in the T dimension
available per firm?  Min, mean, median, that sort of thing?  2-way
clustering hasn't been much explored in the literature yet, but maybe
others can comment if you provide more detail.

--Mark

> The full reference: Petersen (2009) ''Estimating standard 
> errors in finance panel data sets: Comparing approaches'', 
> Review of financial studies, pp 458-459.
> 
> ----------------------------------------
> > Subject: RE: st: RE: Cluster standard errors by time and firm
> > Date: Mon, 11 Jun 2012 16:42:45 +0100
> > From: M.E.Schaffer@hw.ac.uk
> > To: statalist@hsphsun2.harvard.edu
> >
> > Sebastiaan,
> >
> > > -----Original Message-----
> > > From: owner-statalist@hsphsun2.harvard.edu
> > > [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of sebas 
> > > nicaise
> > > Sent: Monday, June 11, 2012 3:52 PM
> > > To: statalist@hsphsun2.harvard.edu
> > > Subject: RE: st: RE: Cluster standard errors by time and firm
> > >
> > > My bad,I was refering to Petersen (2009) ''Estimating standard 
> > > errors in finance panel data sets: Comparing approaches''
> >
> > As Maarten says, you should provide a full reference - not 
> just a title.
> >
> > > Where he argues that to control for serial correlation in 
> estimating 
> > > SEs, they should be clustered by firm and quarter.
> >
> > Do you (does he) mean clustering by "firm-quarters"? Or 2-way 
> > clustering by firm and quarter?
> >
> > Taken at face value, the recommendation does not make sense, since 
> > clustering by firm automatically accommodates within-firm serial 
> > correlation. You have 7000 firms, so the asymptotics required for 
> > clustering on firm alone shouldn't be a problem.
> >
> > And you have only 40 time periods (quarters?) and you are using 
> > 4-quarter differences, so the asymptotics needed for clustering on 
> > time alone, or as part of 2-way clustering, probably would 
> be a problem.
> >
> > My guess is that Petersen's recommendation is either 
> something else, 
> > or is a recommendation for dealing with a different or 
> expanded problem.
> > For example, it might be a recommendation to deal with serial 
> > correlation + contemporaneous correlations across firms. That is a 
> > problem that 2-way clustering can address, but since you don't have 
> > very many observations in the T dimension it could be problematic.
> >
> > --Mark
> >
> > >
> > > My apologies :)
> > > ----------------------------------------
> > > > Date: Mon, 11 Jun 2012 16:40:41 +0200
> > > > Subject: Re: st: RE: Cluster standard errors by time and firm
> > > > From: maartenlbuis@gmail.com
> > > > To: statalist@hsphsun2.harvard.edu
> > > >
> > > > On Mon, Jun 11, 2012 at 4:35 PM, Maarten Buis wrote:
> > > > > You obviously should also read Cox (2012).
> > > >
> > > > This is deliberately obscure to illustrate the point that it is 
> > > > necessary to give full references in order for an 
> > > > interdisciplinary list like statalist to function.
> > > >
> > > > It is the Statalist FAQ that was meant with Cox (2012). 
> To quote 
> > > > the Statalist FAQ (a link is given at the bottom of 
> every post to
> > > > statalist): "Precise literature references please! 
> Please do not 
> > > > assume that the literature familiar to you is familiar to
> > > all members
> > > > of Statalist. Do not refer to publications with just minimal 
> > > > details (e.g., author and date). Questions of the form 
> "Has anyone
> > > implemented
> > > > the heteroscedasticity under a full moon test of Sue, 
> Grabbit, and 
> > > > Runne (1989)?" admittedly divide the world. Anyone who has not 
> > > > heard of the said test would not be helped by the full 
> reference 
> > > > to answer the question, but they might well appreciate 
> the full reference."
> > > >
> > > > -- Maarten
> > > >
> > > > --------------------------
> > > > Maarten L. Buis
> > > > Institut fuer Soziologie
> > > > Universitaet Tuebingen
> > > > Wilhelmstrasse 36
> > > > 72074 Tuebingen
> > > > Germany
> > > >
> > > >
> > > > http://www.maartenbuis.nl
> > > > --------------------------
> > > >
> > > > *
> > > > * 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/
> > >
> > > *
> > > * 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/
> > >
> >
> >
> > --
> > Heriot-Watt University is the Sunday Times Scottish 
> University of the 
> > Year 2011-2012
> >
> > Heriot-Watt University is a Scottish charity registered 
> under charity 
> > number SC000278.
> >
> >
> > *
> > * 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/ 		 	
>    		  
> *
> *   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/
> 


-- 
Heriot-Watt University is the Sunday Times
Scottish University of the Year 2011-2012

Heriot-Watt University is a Scottish charity
registered under charity number SC000278.


*
*   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/