Clustering standard errors at the firm level

[Ralph Klijn]

Dear users,

I am using an unbalanced panel data set with annual data ranging from 1991 to 2012. Moreover, I use Stata 15.

Following a colleague's advice, I have clustered all standard errors at the firm level in my regressions.

What are the exact consequences of doing this? How should I interpret it?

Thank you in advance.

Best regards,
Raphaël

[Carlo Lazzaro]

Ralph:
assuming that you're using -xtreg-, -robust- as well as -cluster- options do the same job: taking heteroskedasticity and/or autocorrelation into account.

[Dimitriy V. Masterov]

By doing this, you are taking into account the fact that when you see the same firm twice in your data this is less information that seeing two distinct firms since the firm's outcome is correlated over time. Typically, clustering will cause your standard errors to increase.

Cameron & Miller's JHR paper would be a worthwhile read.