Heteroskedasticity & serial correlation in panel data with fixed effects

[Jol Blanche]

Hello everyone and thank you for a very helpful forum!

I have roamed the forum and can't seem to find definite answers to my questions.

I am currently writing my thesis about trade flows using the gravity model on panel data with fixed effects using xtreg in stata.
I have 50 panels (pair-trade flows), 6 years each and thus 300 observations in total.

I have tested for heteroskedasticity and autocorrelation and found that my model is subject to both. From what I have been able to find on this forum and through other resources, I can correct for both at the same time by using the vce(robust) command (the fixed effects are at trading states pair-level and thus clustered on this level). Is this correct? I have come across answers to similar questions stating that "if you are dealing with a large N, small T panel dataset.." and "If you're dealing with a large N, small T panel dataset (...) and you suspect autocorrelation and/or heteroskedasticity, you should simply robustifying/clustering your standard errors..." and similar. The question is then what are the tresholds for large N and small T? As mentioned, I have 50 panels * 6 years = 300 observations.

Thus:
1) Does vce(robust) correct standard errors for both heteroskedasticity and serial correlation in this context?
2) What are the tresholds for "large N" & "small T" in this context?

Thank you in advance!

Best regards,
Jol

[Carlo Lazzaro]

Jol:
welcome to the list.
A1) Yes.
A2) I'm not aware of such thresholds (if any exists).
As an aside, please note that robustified/clustered standard errors do the same job under -xtreg-, but are different beasts under -regress-.

[Jol Blanche]

Hi Carlo,

Thank you for the welcome and for your quick response!
It was actually you who I quoted regarding "large N" & "small T", please see the following threads:
http://www.statalist.org/forums/foru...nel-data-model
http://www.statalist.org/forums/foru...-versus-robust

Does these Ns & Ts concern a different setting, or have I misinterpreted it completely?

Best regards,
Jol

[Carlo Lazzaro]

Jol:
you can be sure that you have a small N, large T panel dataset when the time-series dimension exceeeds the cross-sectional one.
That is, you have 10 panelid and 20 waves of data.
It's like to have a rectangular array of data which stands on the shortest side (large N, small T) or on the largest one (small N, large T).
And yes, you're right: my reply #6 http://www.statalist.org/forums/foru...-versus-robust, is miswritten: -xtreg- applies to large N, small T panel datasets (as in your case) , whereas -xtpcse- and the like are advised for small N, large T panel datasets. Sorry for the mishap.

[Jol Blanche]

Carlo,

I see! Thank you for the help!

Best regards,
Jol