Heteroskedasticity Robust standard errors

[arjen19]

Hi there

I have a question concerning the usage of Heteroskedasticity Robust standard errors.

I am analyzing an unbalanced panel data set (835 cross sections, T=3, number of observations 2448, 11 explanatory cross-section specific variables). I further have integrated a time period dummy by clicking the applicable option.

I would like to correct my outputs (regular OLS) for Heteroskedasticity (and if possible for autocorrelation).

Which robust standard errors should I use from the following?
-White diagonal standard errors & covariance
- White cross-section standard errors & covariance
- Cross-section SUR (PCSE) standard errors & covariance

Or are there any other instruments to eliminate the issues of Heteroskedasticity within EVIEWS? Autocorrelation could be eliminated by applying fixed effects (no feasible due to large loss of degree of freedoms T is too small) and endogenously lagging the dependent variable is not senseful. Therefore, these technics are not applicable in my case.

Thank you very much for your support!

Arjen

[EViews Glenn]

White period handles clustering by cross-section, which will lead to correlation between units in a cross-section. These standard errors will be robust to this kind of clustered correlation. See Wooldridge, among others, for details. Alternately, you can explicitly add AR terms to your regression, which will estimate a model assuming a particular form of serial correlation (which may or may not be the kind you want0>

[arjen19]

Thank you very much for your respond.

Just to summarise what you've written. I can apply White period effects in order to get standard errors robust to autocorrelation. Further, I can implement White diagonal effects in order to get heterskedasticity robust standard errors or is White period also robust to heteroskedasticity? Does Eviews offer standard error correction generally referred as heterskedasticity-and-autocorrelation-consistent (HAC) standard errors? (also called Hansen-White standard errors)

Thank you for a short answer on this matter.

Regards Arjen

[EViews Glenn]

White period (cluster by cross-section) error estimates are heteroskedastic and cluster robust so that they allow for E(u_it u_is) to be non-zero for t<>s and to differ across periods. Note that this is a more general correlation structure than the usual form of an AR model. Note also that this estimator requires N asymptotics.

EViews doesn't offer built-in HAC standard errors in the standard panel setting.

The White-diagonal is restricts E(u_it u_is) to be zero for t<>s.