Adjusting standard errors for cross-sectional dependence in Fixed-Effect Panel

[Karl Lamber]

I am currently running a fixed-effect panel regression. First I declared the dataset as panel using ( xtset ID OBS) where ID is 500 firms; OBS is 125 monthly observations. Therefore, in total 62,500 observations.

My goal is to run the fixed-effect regression while adjusting standard errors for cross-sectional dependence. Will running the following regression adjusts standard errors for cross-sectional dependence:

xtreg y x1 x2 x3 x4 x5 i.OBS, fe

I appreciate any help you can provide.

[Clyde Schechter]

Not quite. You also want to specify the cluster-robust standard error. You can do that by adding -vce(cluster ID)- to your -xtreg- command.

(You can also do this by just specifying -vce(robust)-, but the two approaches are synonymous only for -xtreg, fe-. For other panel data commands, these two options lead to different results and only those with -vce(cluster ID)- are properly aadjusted for cross-sectional dependence.)

[Karl Lamber]

Clyde Schechter Thank you for the help. After doing lots of research, I thought clustering by ID -vce(cluster ID)- would adjust standard errors for time-series correlation not cross-sectional dependence. Thanks for clarifying though, you've been helpful.

[JanDitzen]

Hi Karl,
depending on the model, you might want to have a look at Driscoll-Kraay standard errors as outlined in Driscoll, Kraay (1998, The Review of Economics and Statistics) or the common correlated effects approach by Pesaran (2006, Econometrica). For both papers commands in Stata exists such as xtscc or xtdcce2. Also it is important to test for cross sectional dependence, see Pesaran (2015, Econometric Reviews) and xtcd, xtcsd and xtcd2 in Stata.
Hope this helps!