st: fixed effects with error autocorrelation coefficient that varies across larger units that panels nest into?

[Adam Storeygard <adam.storeygard@tufts.edu>]

Dear Statalist,

Is there a Stata command (ideally accessible in version 12.1) 
implementing a fixed effects (within) estimator of a panel model with 
AR(1) disturbances such that the autocorrelation coefficient is allowed 
to vary across larger units that the panel units nest into?

Somewhat more formally, I would like to estimate:

y_it = a + x_it * B + u_i + e_it
where
e_it = rho_c * e_i,t-1 + z_it
and z_it is independent and identically distributed. i indexes 
individuals, c indexes cities (individuals nest within cities), and t 
indexes years. I would be running this on a balanced panel of 100 to 
1000 individuals (nested in 5 to 50 cities) for 10 to 20 years.

-xtregar- (version 1.5.2  27jan2012) would be suitable except that it 
does not, as far as I can tell, allow the autocorrelation coefficient 
rho to vary by city. -xtgls- (version 2.4.2  03feb2011) has the 
-corr(psar1)- option allowing the autocorrelation coefficient to vary by 
individual, which does not impose as much structure as I would like, and 
is not a fixed effects (within) estimator.

Thank you for your consideration,

Adam
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/