st: Re: Newey-West robust errors

[Kit Baum <baum@bc.edu>]

Bruno,

It is the case that if the only departure from iid errors is positive  
AR(1), the OLS standard errors will be downward biased relative to the  
correct standard errors. But in the presence of heteroskedasticity and  
autocorrelation, all we know is that the OLS standard errors are  
biased and inconsistent. They could be higher or lower than the HAC  
standard errors. A serious issue, though, is that tests for non-iid  
errors are sensitive to the maintained hypothesis that the model is  
specified properly. Residuals will be correlated, for instance, if an  
important variable is omitted from the model (in your case, y(-2)?) So  
consider those test results as a possible signal of misspecification.  
Use the RESET test (estat ovtest) and see what it says.

Best wishes
Kit

Kit Baum, Boston College Economics and DIW Berlin
http://ideas.repec.org/e/pba1.html
An Introduction to Modern Econometrics Using Stata:
http://www.stata-press.com/books/imeus.html


On Feb 20, 2009, at 22:28 , Bruno Schroder wrote:

> Dear Professor Baum,
> I`m an undergraduate student majoring in economics and have visited  
> your awsome webpage recently searching for an answer. Is it possible  
> that after correcting for autocorrelation in residuals the new  
> standard errors get smaller than the usual OLS ones?
> It`s all about because I have running a regression of "y on y(-1)  
> plus controls" (annual data) and through OLS many of the parameters  
> are not statistical significant. Then, I`ve run tests for  
> heteroskedasticity and autocorrelation, always rejecting the null.  
> So I do not need to correct for these "problems", actually. But when  
> I use HAC Standard Errors, those parameters become significant. Is  
> there any problem?
> I appreciate your comments.
> Best regards,
> Bruno Schroder

*
*   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/