Hi all--
I just upgraded from Stata 9 to Stata 10 and have been re-running some of my old .do files. One in particular perplexes me.
I have an unbalanced time-series cross-section model of the following sort:
y = x1 x2 y_t-1 D Counter
Where D are fixed effects and Counter is a variable that adds 1 for every year (1 in year 1, 2 in year 2, etc.). My t/n is about 1/2 (n=19, t=10) so I'm in uncharted territory here. As a robustness test given the relatively short panel and the LDV with fixed effects, I've been estimating an Arellano-Bond model using xtabond.
xtabond y x1 x2 Counter
In Stata 9, the variable Counter turns into a column of 1s due to differencing, which makes it colinear with the constant so Stata drops it, as I would expect.
But in Stata 10, using the exact same commands, I get an estimate of the effect of counter on y. It's even significant.
My questions are 2:
1. How could this be possible? I've noticed that in Stata 9, the output from xtabond takes the form of differenced independent variables plus a lagged differenced independent variable, but in Stata 10 output does not include differences. Does this mean that xtabond in Stata 10 does not estimate the differenced Arellano-Bond panel data estimator that it did in Stata 9?
2. I'm also exploring the Arellano-Bover/Blundell-Bond estimator using xtdpdsys (this was the reason I upgraded to Stata 10). I get an effect for Counter in this model as well. Does this mean that this model does not estimate differences, but rather levels? Wouldn't my Counter variable drop out of the instrument set (as I understand it, xtdpdsys uses differences to instrument levels)?
Thanks for your help!
TP
*
* 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/
