asked 2014-01-06 09:02:00 +0000

Pantera gravatar image

updated 2014-01-08 08:38:07 +0000

Hi all,

To eliminate or reduce the influence of potential autocorrelated and heteroscedastic residuals and to obtain unbiased standard errors, it is possible to use HAC (Newey-West) variance estimator. “Unfortunately”, I must say, there are a lot of options described in the literature with regard to choosing the “right” kernel method (Bartlett, Quadratic spectral, etc.) and bandwidth (Andrews automatic, Newey-West automatic etc.). However, so far I do not find any unambiguous method to use. It is also frustrating because different HAC-estimators seem to give different results. Question: What is regarded as the “best”, most robust or safe method to use? Any suggestions?

edit retag flag offensive close merge delete


Each method has it's own advantages and disadvantages subject to the data and what is important to the modeller. Potentially someone has reviewed these and/or one of the graduate manuals may cover these points.

SHAZAMHelp gravatar imageSHAZAMHelp ( 2014-01-10 18:00:36 +0000 )edit

Thank, I'll check the literature.

Pantera gravatar imagePantera ( 2014-01-15 12:57:08 +0000 )edit