How to Adjust Standard Errors for time-series correlation and cross-sectional dependence.

[Karl Lamber]

I have a large dataset currently set up as a panel, (1000 Firms and 200 Months). I would like to run a fixed-effect panel regression and adjust standard errors for both cross-sectional and time series correlation as my data suffers from both issues.

First, I tried - cluster(ID) - but it seems to only correct for time-series autocorrelation. Then I clustered by(ID) and added a time dummy, but my R-Squared increased dramatically.

I was wondering if it is possible to cluster by both ID and Time? Also, why is the R-Squared increase significantly when I include a time-dummy?



Your help is much appreciated as I am a beginner Stata user.

[River Huang]

Please first "ssc install reghdfe", and then "help reghdfe". The -reghdfe- command can account for both issues you encountered. For instance,

webuse nlswork, clear

reghdfe ln_w grade age ttl_exp tenure not_smsa south , absorb(idcode year)

// cluster at both dimensions
reghdfe ln_w grade age ttl_exp tenure not_smsa south , absorb(idcode year) vce(cluster idcode year)

[Karl Lamber]

River Huang Thank you very much.

[Joao Santos Silva]

Dear Karl Lamber,

You may also want to check the Driscoll-Kraay standard errors computed by -xtscc-

Best wishes,

Joao

[Victoria Consolvo]

Will clustering on the fixed and time effects account for serial correlation in the same manner as xtscc accounts for it?