clustered robust standard errors

[RheaMaria]

Hi,

Could someone please tell me how to implement clustered robust standard errors in eviews 7 as I do not know how. Would really appreciate the help.

Best
Rhea

[EViews Glenn]

http://forums.eviews.com/viewtopic.php? ... ster#p1191

[RheaMaria]

Hi Glenn,

Thanks for your post. If I understand correctly, I should set up the panel data with the cluster variable as the cross section dimension? I have done so as I want to cluster by Country and Country is my cross section dimension for my fixed effects regression.

What I am concerned about is a very high R squared (0.998) which suggests autocorrelation and serial correlation.

I am very new to eviews so am not sure how to control for autocorrelation/heterskedasticity/serial correlation in panel data in Eviews 7.

Would really appreciate your help here.

Thanks
Rhea

[EViews Glenn]

Yes. Cross-section by county.
The R2 that we report includes that obtained from the fixed effects. I rarely look at R2 in general, and wouldn't worry about it here.
The White period will compute standard errors that are robust to within cluster correlation. Note that the standard errors are asymptotic in the number of clusters.

[RheaMaria]

Thanks for your reply Glen - it was really helpful.

I was wondering if the DW statistic is useful in determining the continued existence of autocorrelation in my model? When I put a common AR(1) term in to improve it from 1.56, it moves up to 1.82 but the residual graphs stay the same.

Would really appreciate some guidance.

Thanks
Rhea

[EViews Glenn]

It is a particular test, though not the preferred in the literature. We don't have anything built-in, but if you check Baltagi (1998), there are discussions of various tests that are probably not that hard to compute in EViews.

[RheaMaria]

Hi Glen,

Thanks for your response. Just to be clear, using the White Period option with the cross-section variable as my cross section ID produces clustered robust standard errors taht are robust to within-cluster correlation. THat means that there still needs to be testing for between cluster correlation and hence further corrections could be made i.e. adding a common AR(1) term to correct for this?

[EViews Glenn]

The first sentence is correct. As is the first part of the second. The AR(1) would not be the ordinary approach to correcting for contemporaneous between cluster correlation as it is primarily concerned with a particular form of within cluster correlation, the commonality of the AR(1) coefficient notwithstanding.