Question about standard error clustering in a DiD-analysis

[Olivier Van Coppenolle]

Hello everyone,

For our master’s thesis we are conducting a difference in difference research which researches the effect of Brexit on the ROA of banks in the EU.
We have a panel dataset with 79 banks and data on ROA/ROE and control variables from 2013 to 2019.
We indicated the panel structure in stata using xtset BANK YEAR.
When conducting some regressions to measure what the coefficients and standard errors of the DID term were, we noted something strange:

In the table you can find the standard errors of the DID term with the level on which these se’s are clustered.

I don’t understand why the standard errors of clustering of bank level is the same as robust se’s and why the country clustering is the same as country x bank clustering?
Is this some kind of fault and how can we correct this?
I also provided the commands used in stata and a glimpse of our dataset to show you how it’s structured.
Thanks in advance!

Olivier

[Andrew Musau]

On the meaning of robust standard errors in -xt- commands, read the technical note on page 15 of xtreg's PDF manual entry: https://www.stata.com/manuals13/xtxtreg.pdf. For #2, if banks are completely nested in countries (i.e., no bank is registered in different countries across years), then bank clusters are redundant if you want to have country clusters as well. Anyway, there does not seem to be strong theoretical reasons to experiment with combinations of clusters. If the policy was implemented at the bank level, just stick with -robust-.