Hi all:

I am using xtabond2 to estimate system GMM. My main interest variable is not lagged dependent variable, but a categorical endogenous independent variable (X). I include this endogenous variable in gmmstyle (i.X, laglimits(2 4)) My predetermined variables include some binary variables, such as whether get married, and whether living with children. I include them in gmmstyle (i.marriage i.child, lag(1 3)). However, I am not sure whether I can use system GMM if binary/categorical variables are endogenous variables or predetermined variables? My concern is IV matrix created for dummy variables may contains lots of zeros, which may cause weak instruments or violate results.

Another question I want to ask is the coefficients of fixed effect model and system GMM. I noticed some coefficients of system GMM in my model is two times of coefficient in FE model. I do not know whether the coefficients are comparable between FE and GMM?

The following is part of my results. My endogenous variable is a categorical variable, including 7 types. I show the results of FE and GMM with different numbers of instruments. In the 6th type (second to last row), the coefficient of FE is 0.0928, while in GMM, such coefficient becomes 0.2060 or lager.

Code:
FE Full instruments 2-4 lags 2-3 lags
Lag dependent -0.1553 0.1388 0.1467 0.1618
0.0554 0.1000 0.0943 0.1025
Endogenous categorical variable
1 -0.0042 0.1176 0.1215 0.0841
0.0269 0.0804 0.0790 0.0797
2 0.0588 0.0390 0.0996 0.1155
0.0294 0.0741 0.0759 0.0868
3 0.0696 0.0988 0.1328 0.1440
0.0369 0.0750 0.0983 0.1161
4 0.0909 0.1660 0.1893 0.1899
0.0312 0.0756 0.0782 0.0856
5 0.0371 0.1791 0.2329 0.2178
0.0361 0.0982 0.1100 0.1054
6 0.0928 0.2060 0.2573 0.2285
0.0381 0.0994 0.0996 0.0976
Thank you in advance!

Connie