re: st: Arbitrary serial correlation in the error term

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

<>
Ari said

Can somebody give =A0a definition of =A0"arbitrary" serial correlation?=
 Is
it related in any way to AR(1), MA(1)? How?

A HAC estimator of the VCE produces estimates robust to arbitrary departures from iid. An implementation such as Newey-West requires that you specify the maximum order (lag) at which you think autocovariances of the error process are nontrivial. Given that, the autocovariance function of the error process may take on any form within that finite lag order that corresponds to a stationary process.

The SSC routine -ivactest- by Baum and Schaffer will allow you to test (for OLS regression as well as IV/2SLS) that certain autocorrelation coefficients of the error process are nonzero: for instance, to test that rho(2,3,4) are jointly zero, while allowing rho(1) to be nonzero. That might allow you to reduce the number of lags used in a Newey-West estimator. Note, however, that using too few lags to capture the non-iid nature of the errors will yield non-robust (and inconsistent) estimates of the VCE.

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/