Newey-West specifications

[oksanakim]

Hello All

Please advise on the following. When estimating an OLS regression, I am using the HAC robust standard errors. I realized the HAC estimation has a variety of options under the "HAC option" button, and those affect the estimation output significantly. For example, the default options suggested by EViews is:

Lag spec: none
Kernel: Bartlet
Bandwidth: NW-fixed

Now, if I change the Kernel option to "None" and remove the "NW-fixed" from the bandwidth field, I ended up with significantly different standard errors and t-stats, compared to the default Eviews output. Could you advise when it would be reasonable to change the EViews default options and specify "none" in the kernel and bandwidth field. The reason I am asking is because the estimation output is highly sensitive to choice of options in the Kernel and Bandwidth fields. Isn't better to specify "none" in those fileds rather than guess on the most appropriate specification should be? Thank you.

[oksanakim]

All right, I just realized that if I specify "none" in the kernel and bandwidth fields, then the estimation is no longer HAC robust se, it's equivalent to White estimation.

[oksanakim]

A follow up question then. When does it make sense to use "use-specified" in the bandwidth field? And how would I determine a number of lags in this case? Since the output is affected by the choice of lags and also changing the bandwidth option. Thanks.