Correction for serial correlation in panel data

[James Hallam]

Dear all,

I am working on a panel set containing 25 countries over 15 years and when using the 'xtserial' I find serial correlation in the model and I plan to do a simple linear model using pooled OLS, fixed effects and random effects. I have a few questions if anyone would be so kind to respond. Firstly, how would I correct the fixed effects model for this serial correlation? I don't think its going to be enough to just correct the standard errors and when I use the linear regression with AR(1) disturbances option on stata I lose significance in every variable in the model! Furthermore, given the serial correlation how would I conduct I robust hausman test? Any help would be much appreciated.

Thanks!

[Carlo Lazzaro]

James:
welcome to the list.
I can't follow you about performing pooled OLS and -xtreg, fe- and, again, -xtreg, re-; serial correlation in itself does not rule out any of these approaches; besides, the literature in your research field should give you some hints about the method usually followed by others presented with the same research topic;
- clustering SEs is actually the way to fix both heteroskedasticity and serial correlation under -xtreg-;
- for robust hausman test you can take a look at http://www.stata.com/statalist/archi.../msg01069.html.

As a closing-out remark, your chances of getting helpful replies are conditional on posting what you typed and what Stata gave you back, via CODE delimiters (these topics are covered in FAQ, that every lister is asked to read before her/his first posting). Thanks.

[Phil Bromiley]

You might also consider xtregar. While clustering takes care of the SEs, I think that theoretically a method that considers the serial correlation in the beta estimates should be more efficient.