cluster on firm or two-way level?

[Johnny Cash]

Hello all,

If the model is y_it = x1_it + x2_it + x3_t + e_it, which means there is a time-serial variable in the panel regression model. How to choose the cluster level for standard error?

Should I cluster on firm level or two-way level ( firm and time level)?

Thank you!

[Carlo Lazzaro]

Johnny:
you can go along the following lines (I assume that you start with the -fe- specification):

xtset firm time
xtreg <depvar> <other_predictors> i.time <controls>, fe
testparm i.time

-testparm- will inform you about the joint statistical significance of your -timevar-.

Again under the -fe- specification, you may want to consider the community-contributed module -reghdfe- (that allows multway -fe- to be collected).
As an aside, cluster should be made on -panelid-.