Python：PanelOLS-双向群集？

Question

In Python/Pandas I use the PanelOLS function. This function gives you the ability to cluster your standard errors. For instance:

PanelOLS(y=panel.Y, x=panel[['X1', 'X2'], nw_lags=10, time_effects=True,
         cluster='time')

But I would like to also cluster by standard errors by entity as well as in time .

Is there a way to do so? If not, what about the panel function in statsmodel ? I have a hard time to find the documentation on Panel regression using statsmodel.

UPDATE If I control using the newey-west lags ( nw_lags ) , isn't this like cluster='entity' ? So, if I use both nw_lags and cluster=time , it is like doing a 2-way clustering?

Answer 1

In Thompson (2011) "Simple Formulas for standard errors that cluster by both frim and time" he describes how you can achieve double-clustering with commands that allow only one way clustering.

Denote your variance estimator clustered by entity by V_ent and your variance estimator clustered by time by V_time and the heteroskedasticity-robust estimator by V_white . Then the double clustering standard error is given by V = V_ent + V_time - V_white . So you can use your PanelOLS regression to give you each of these variance estimators and then compute the double-clustering standard errors by yourself.