Fixed Effects Regression with few clusters

[Juan Deto]

Hi all,

I have a balanced panel from Apr2021 to June2022 in hourly data.

xtset token_id date

Token_id can have three different values (1,2,3) and date is just the date from nov to juli in hours. I have to do a fixed effects regression and also check for heteroscedasticity but I am not sure how.
my current regression:

xtreg rel_pd abs_future_return abs_underlying_return volume_underlying volume_perp future_spread_CS vcrix, fe vce(robust)

I am pretty sure that the std error are not right. I already researched and found "A Practitioner’s Guide to Cluster-Robust Inference" from Cameron and Miller but could not quite grasp what I have to do precisely.

Thanks in advance!

[Carlo Lazzaro]

Juan:
the advice is simple here: just forget the surely misleading non-default standard errors due to the negligible number of your clusters.
In addition, I would be much more concerned about the correct specification of your regression models: you clearly have a T>N panel dataset and -xtreg- cannot help you here.
You should be better consider -xtgls- or -xtregar-.

[Maxence Morlet]

Carlo is absolutely right. Your ratio T/N practically tends to close enough to infinity for quite a few desirable asymptotic properties of several covariance-matrix estimators to kick-in.

In addition to Carlo's insights, you may want to check out Hoechle's (2007)

ssc install xtscc