xtabond2 in an unbalanced panel with time FE

[Michael Gmeiner]

I have an unbalanced panel at the quarter level and want to make sure I'm implementing xtabond2 correctly. I'm only interested in the effect of x on y and believe it is exogenous, I just need to deal with the lagged outcome variable on the right hand side in a setting with fixed effects and small T.

This,
https://www.statalist.org/forums/for...m-time-dummies
implies that time fixed effects should only be in the level equation, and that the iv() option should always be used with the eq() suboption.

What I'm running is the following, are there any suggestions for why this may not be the best method to estimate? Is there anything I need to do to account for having an unbalanced panel?

xtabond2 y x l.y i.time ,iv(x, eq(diff)) iv(x i.time, eq(level)) gmm(l.y, lag(1 5)) small cluster(region)

[Sebastian Kripfganz]

There is a problem in the current version of xtabond2 when time dummies are specified with the factor variable notation. Some time dummies will be dropped (listed in the regression output as omitted), as they should. While the coefficient estimates and standard errors are fine, the degrees of freedom for the overidentification tests might be incorrectly computed in the presence of such omitted dummy variables. I recommend to manually create the dummies first and only add the dummies that are not omitted to the variable and instruments lists. Alternatively, you could use the xtdpdgmm command with the option teffects.