Ivan Tasic <[email protected]> wrote:
> Howdy!
>
> I have three questions. The basic model used for -xtlsdvc- is as follows:
> y_it = gamma*y_i,t-1 + x'_it*beta + eta_i + e_it
>
> 1. Is there any way to specify how many lags of dependent variable we want
> to use in -xtlsdvc-, or one lag is the only option? What's the consequence
> of including additional lags manually - as regressors?
No, the bias correction in -xtlsdvc- can only work for a model with one
lag of the dependent variable
> 2. Would bias correction work properly if we use time dummies?
Yes, it will.
> 3. I use a large TSCS (n=67, T=398, weekly data), and the bias looks very
> small compared to -xtreg, fe-. I guess that asymptotics kick in pretty well.
> Since it took Stata 9 forever to calculate robust standard errors (I did use
> the highest level of precision O(1/NT^2), is there any advice how to speed
> it up, except to buy a better computer?
Just be patient... In your case the time dimension seems larger than
the typical sample sizes for which these procedures are designed and tested.
Anyway, -xtlsdvc- is much faster when initialized by the Anderson-Hsiao
estimator with no apparent loss compared to the other two initial estimators.
Also, keep the number of replications to a reasonable level (right now
I'm using vcov(50) with N=31 and T=24). For more explanation and references
look at my paper
http://ideas.repec.org/p/cri/cespri/wp165.html
Hope this helps
Giovanni
(author of -xtlsdvc-)
--
Giovanni S.F. Bruno
http://ideas.repec.org/e/pbr136.html
Istituto di Economia Politica, Universit� Bocconi
Via U. Gobbi, 5, 20136 Milano
Italy
tel. + 02 5836 5411
fax. + 02 5836 5438
*
* 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/