Heteroskedasticity / Autocorrelation in Panel Data

[Rick Last]

Hi all,

I have a question with regards to the presence of heteroskedasdicity and/or serial correlation in my panel data set. I am using a fixed effects estimation (fe).

I tested for heteroskedasticity with the xttest3 command and it indicates there is evidence for heteroskedasticity (p-value = 0.000...)

I tested for serial correlation with xtserial and it indicates ass well evidence for serial correlation (p-value=0.000...)

Is is sufficient for me to use robust standard errors (vce(robust) to avoid problems with hetereskedasticity/serial correlation? I red also something about vce(cluster id) st errors, but using these two gives me identical results.

I also red something about that with a small number of time periods heterskedastic robust st. errors are not valid anyway. My number of time periods is 15, and the number of cross sectional entities is approx. 4800.

If someone could give me some advise about whether which st. errors are sufficient I would be verry happy. On the internet I red some contradicting opinions which made me a bit confused.

Rick

[Rick Last]

I also red something about Driscoll Kraay standard errors, which can be used with using the xtscc command instead of xtreg. Can anyone tell me if I this is the right method to deal with heteroskedasticity and serial correlation with my panel data? Or are robust standard errors still better?

I would really appreciate any help

Thanks

[Carlo Lazzaro]

Rick:
if you're dealing with a large N, small T panel dataset, you can use -xtreg,fe- with clustered/robustified standard errors (as you found out yourself, they do the same job under -xtreg-, unlike under -regress-) to manage both autocorrelation and heteroskedasticity.

[Rick Last]

Carlo,

Thank u very much for the confirmation!