Newey-West Standard Errors

noetsi

Fortran must die
#1
Can these be generated in Proc AutoReg ? They exist in PROC REG (which makes little sense for time series regression to me] and PROC MODEL which I do not know. I was hoping they exist in PROC AUTOREG, but if they do I can not find them.
 

Dason

Ambassador to the humans
#3
Wouldn't part of the point of using autoreg be so that the errors aren't autocorrelated? Since the model already accounts for the autocorrelation you don't need to use an estimator whose goal is to account for autocorrelation remaining in the error term.
 

ondansetron

TS Contributor
#4
Yeah, I'm not sure it's appropriate in an autoregressive model. Half those fancy remedies are for the econometricians who want to bulldoze through with OLS.
 

noetsi

Fortran must die
#5
Proc Autoreg will address autocorrelation if you get the lags correct - but that is far from certain. Newey-West would address the problem even if you don't I would think (but I have not seen this addressed in honesty).
 

noetsi

Fortran must die
#6
It turns out PROC AUTOREG does do this after all (and this deals with hetero which is why they are called Newey West HAC sometimes

PROC AUTOREG ....
Model..../COVEST =NeweyWest;