re: st: Issues with xtabond2

[Christopher Baum <kit.baum@bc.edu>]

<>
Eric said

I just came back to look at your post more closely. With 132 observations and the  gmm option without any limit on the number of lags, you are generating a huge number of instruments. This is creating numerical problems. Even if it were not, this affects the staistical efficiency of your results. You may want to try fixed effects, relying on large T theory.


Quite so. The rationale for A-B or Anderson-Hsiao is the mitigation of Nickell bias, which can be severe with T<10, as it is a function of 1/T. As T gets large, Nickell bias ls likely to become inconsequential, so that a standard FE approach (or IV-FE, given that you may want to consider some other regresssors endogenous) would probably suffice.

Kit


Kit Baum   |   Boston College Economics & DIW Berlin   |   http://ideas.repec.org/e/pba1.html
                             An Introduction to Stata Programming  |   http://www.stata-press.com/books/isp.html
  An Introduction to Modern Econometrics Using Stata  |   http://www.stata-press.com/books/imeus.html


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