Dear statalisters:
1. I am looking for a test for higher-order serial correlation (HOSC) in a
dynamic FE model with unbalanced panels (N=11 groups & T=50 years;
unbalanced).
Due to ideosyncratic structure and unequal spacing of my independent
variables (state elections votes, which occur roughly every 3-5 years), I
expect there may be third/fourth/fifth order SC within panels, so a simple
AR1 correction (Prais Winston etc.) won't do the job. Moreover, the uneven
spacing of time points renders familiar tests for SC, such as
Durbin-Watson's d or Breusch-Pagan-LM tests inappropriate. Arellano-Bond
tests ("abar") would work in principle to pick off HOSC but are not
appropriate for fixed-effects regressions for dynamic models. Is there any
viable alternative?
2. Alternativly, is there an effective way to correct for (panel specific)
HOSC in a dynamic FE model with unbalanced panels?
I have tried to use a specification with HET&SC resistant (Newey) SEs that
allow for SC of unknown form up to a prespecified period of lags. However,
as far as I know Newey SEs are based on a single (pooled) autocorrelation
parameter estimate across panels, and what I ideally want is to allow for
unique serial correlation coeffcients for each panel.
xtgls, corr(psar1) would do this job. Unfortunalty, this again won't correct
for HOSC (only 1st order) and additionally the GLS estimator has very poor
asymptotic properties in my small N, large T context. The GLS SE estimates
seem overly-optimisitic.
I would appreciate any help on this. Thank you very much!
Best,
Jens
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
