Cost of using AC or HAC standard errors in ivreg2 and xtivreg2

[Jan Dithmer]

Dear all,

I am using ivreg2 and xtivreg2 for estimation of a linear panel data model. Available serial correlation tests suggest that there is no problem with firs oder higher-order autocorrelation. Therefore, I just use the robust option together with gmm2s and cluster(id) in ivreg2 or fe gmm2s robust in xtivreg2 .
But, I am wondering if it still makes sense to use autocorrelation-consistent (AC) or heteroskedastic and autocorrelation-consistent (HAC) or, in other words, is there any cost associated with the correction for autocorrelation?

Thank you for your help!

Best, Jan

[Jan Dithmer]

Can anyone help?

[Mark Schaffer]

There's a literature on this. A good place to start is Cushing and McGarvey, "Covariance Matrix Estimation", in Laszlo Matyas (ed), Generalized Method of Moments Estimation, CUP 1999. There's a section in the chapter discussing finite sample performance of HAC estimators. They don't use a panel setting, but still useful.

[Jan Dithmer]

Ok, thanks Mark, i will check the literature you suggested!