Heteroscedasticity and autocorrelation in OLS model

[Nermine Medhioub]

Hello Stata community!
Well, my OLS panel data suffer from Heteroscedasticity and autocorrelation. To the best of my knowledge ( I might be wrong), to correct only for Heteroscedasticity we can add the vce(robust) command. My question is how to correct for both Heteroscedasticity and autocorrelation using OLS (regress) and fixed/random effect (xtreg).

[George Ford]

vce(cluster "id_variable"), if you have enough N in your NxT in your panel. How many you got?

[Nermine Medhioub]

my study covers 39 companies for 3 years

[George Ford]

good to go.

add cluster("your_company_id_variable") to get clustered SE.

[Nermine Medhioub]

thank you so much @George Ford

[Maxence Morlet]

That is not very many clusters, I would use cluster bootstrap with the user written command boottest

[George Ford]

Yeah, you could boottest it for good measure. Easy enough to do and really fast, but the results will be nearly identical with 39 companies. There's a lot of ideas about how many clusters is enough (some saying 10, others saying 50, some say more) and better to be safe than sorry.

[Maxence Morlet]

42, hitchhiker's guide to the galaxy!

[Carlo Lazzaro]

Nermine:
an useful reference on clustering standard errors is Cameron_Miller_JHR_2015_February.pdf (ucdavis.edu)

[George Ford]

Actually, Max, in this case the magic number is 39. 39 is just right.