Robust errors with industry + time clusters

[Gabriele Mauro]

Dear Stata community,

i have a sample of 146 units that are nested within 108 firms and 131 dates, hence the OLS cross-sectional residuals may be non-independent. In the OLS, the parameter estimates are not efficient and this will lead to biased test statistics. I should account for the nested data structure using cluster robust standard errors, clustering for both firm and date (and also heteroskedasticity). A question for you is: what syntax should I use ? reg y x1 x2...., vce(cluster time industry) ... but this does not adjust for heteroskedasticity ... any advice very welcome!

[Maxence Morlet]

What you have used are two-way clustered standard errors (Cameron et al., 2011) which adjust for clustering and heteroscedasticity.

One thing to note is that if there is dependence on a coarser level than the one on which you are clustering, standard errors will be biased downwards (Cameron et al., 2015).