Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: xtpcse vs. newey2 or itivreg2


From   "Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: xtpcse vs. newey2 or itivreg2
Date   Mon, 3 Mar 2008 16:12:51 -0000

Mark

> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of 
> Mark Dincecco
> Sent: Monday, March 03, 2008 3:19 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: st: xtpcse vs. newey2 or itivreg2
> 
> Hello,
> 
> I have an unbalanced panel data set with a very large T to N ratio.
> Essentially, I follow 5 countries over a period of 200 or more years.
> I would like to correct for contemporaneously correlated 
> errors, panel heteroskedasticity, and common serial 
> correlation. I have opted to use xtpcsce, newey2, or xtivreg2 
> rather than xtgls following Beck and Katz (1995).
> 
> The basic specifications look like:
> 
> *xtpcse
> 
> xi: xtpcse y x1 x2, corr(ar1)
> 
> *newey2
> 
> xi: newey2 y x1 x2, force lag(1)
> 
> *xtivreg2
> 
> xi: xtivreg2 y x1 x2, bw(2) robust small
> 
> I have two questions. The first is the most critical:
> 
> 1. Is the only difference here between newey2 or xtivreg2 AND 
> xtpcse that xtpcse controls for contemporaneously correlated 
> errors while newey2 or xtivreg2 does not?

Not quite.  As I understand it, xtpcse uses a specific functional form,
namely AR(1).  xtivreg2 (and newey2) do not.  Rather, they are robust to
arbitrary autocorrelation, but underlying this are asymptotics that
require the number of lags used in the calculation (the bandwidth) to go
off to infinity.

Hayashi's (2000) "Econometrics" textbook has a good discussion.  For a
very abbreviated discussion, see Baum-Schaffer-Stillman in the latest
issue of the SJ (Vol. 7, No. 4, 2007). 

> I ask because newey2 offers better results than xtpcse but I 
> wonder if that is because newey2 does not correct for 
> contemporaneously correlated errors. If not, then I will go 
> with the newey2 estimates.
> 
> 2. Is the only difference here between newey2 AND xtivreg2 
> that xtivreg makes an adjustment for the degrees of freedom?

No, it's the other way around.  newey2 makes a small-sample adjustment
and reports t-stats.  xtivreg2 will make this adjustment if you supply
the -small- option; without it, it reports z-stats and does not make the
adjustment.

Also, xtivreg2 has a wide range of kernels available beyond the Bartlett
kernel ("Newey-West") used by newey2.  There are some other differences
as well (e.g., handling of gaps in time series, which I think xtivreg2
handles more gracefully - at least, Kit Baum and I put a fair amount of
thought into how to handle it in the straight time series context).

Cheers,
Mark (author of xtivreg2)

Prof. Mark E. Schaffer
Director
Centre for Economic Reform and Transformation
Department of Economics
School of Management & Languages
Heriot-Watt University
Edinburgh EH14 4AS  UK
44-131-451-3494 direct
44-131-451-3296 fax
http://www.sml.hw.ac.uk/cert


> I ask simply for completeness.
> 
> Thanks very much for your help. I really appreciate it.
> 
> Sincerely,
> Mark
> 
> PS. This inquiry was first submitted on Sunday. I thank Clive 
> for his suggestion at that time. However, as he notes his 
> answer was insufficient. I thus resumbit an updated version 
> of the questions now.
> *
> *   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/
> 

*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index