Re: st: Robust Standard Errors in Paneldatasets

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

<>
On Oct 26, 2010, at 2:33 AM, Leon wrote:

> Hi, I am new to Stata and try to measure herd behavior as deviations in the return dispersion of a large panel dataset. Hence, I wonder which regression type and which standard errors are most applicable as they should correct for heteroscedasticity and autocorrelation. I recently read these two articles about robust standard errors in panel datasets and can't figure out which SE I should use and in case of the clustered method how to apply this to Stata.
> 
> 
> Petersen, M. A. 2008. Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches. Review of Financial Studies 22:435-80.
> 
> Driscoll, J., & Kraay, A. (1998). CONSISTENT COVARIANCE MATRIX ESTIMATION WITH SPATIALLY DEPENDENT PANEL DATA. Review of Economics & Statistics, 80(4), 549-560.
> 
> I found various methods to apply the regression in Stata and hope you can help me to choose the right one, if any.
> 
> * regression using Driscoll-Kraay SEs (need to install the xtscc package first)
> xtscc depvar varlist, fe
> 
> * regression using Newey-West SEs
> newey depvar varlist, lag('T-1') force
> 
> * regression using White SEs
> xtreg depvar varlist, vce(robust)
> 
> * normal panel regression
> xtreg depvar varlist, fe robust
> 
> * found as well
> ivregress gmm depvar varlist, vce(hac nwest opt) perfect

None of what you have found deals with clustering. xtreg without the fe option is random effects, which is a.s. inappropriate for finance panels. newey and ivregress fail to take the panel nature of the data into account (in fact the ivregress command you give will not run on multiple panels, and the newey with undocumented -force- option is likely to think your data are one long time series). I would look at Schaffer's -xtivreg2-, on SSC, which will allow you to estimate a model with one-way and two-way clustering (see my BOS'10 and UKSUG 2010 presentations, on my RePEc page below). Clustering allows you to deal with arbitrary heteroskedasticity across panels and aribtrary correlation within panels. Two-way clustering also allows you to consider common effects hitting all firms at the same point in time. See the discussion of clustering in Baum/Schaffer/Stillman papers, Stata Journal 3(1) [free] and 7(4), available in preprint form on my website.

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/