Clustering standard errors for heteroskedasticity and firm clustering

[Tomas Milo]

Hello,

I am wondering if someone can confirm that the below code is correct. Specifically, I want to make sure that I am properly adjusting the standard errors based on a paper that I am trying to follow, where the author adjusted standard errors for heteroskedasticity (White 1980) and firm clustering (Petersen 2009).
My code is:

reghdfe dependent_variable variable_of_interest control , absorb(firm year) vce(cluster firm)

Is the above code properly adjusting the standard errors for heteroskedasticity and firm clustering? I read section 20.21.2 from "https://www.stata.com/manuals13/u20.pdf#u20.21Obtainingrobustvarianceestimates" and I am still a bit unsure.

Thank you in advance

[Andrew Musau]

If you have panel data with \(N>>T\), Stock and Watson (2008) show that clustering using the panel identifier results in standard errors that are robust to the presence of heteroskedasticity and arbitrary forms of serial correlation. But you need to have enough clusters, otherwise the bias from a small number of clusters outweighs any benefits from clustering (rule of thumb 30+ clusters are needed). By the way, clustering using the panel identifier is the default for panel data estimators in Stata (xt-prefixed estimation commands) if invoking the -robust- option. See more discussion on this in the PDF manual entry of xtreg.

Reference:
Stock, J. H., and M. W. Watson. 2008. Heteroskedasticity-robust standard errors for fixed effects panel data regression. Econometrica 76: 155–174. https://doi.org/10.1111/j.0012-9682.2008.00821.x.

[Tomas Milo]

Hi Andrew,

Thank you for the response. If I understand correctly then, by using

vce(cluster firm)

, which has about 200 clusters, then I am controlling for both heteroskedasticity and firm clustering?

[Andrew Musau]

Yes, provided

xtset firm year

issues no error message, i.e., you have firm-level panel data.

[Tomas Milo]

Thank you