Xtabond2 and the usage of weights

[Rui Agostinho]

Hello everyone,

I have a doubt regarding the correct usage of weights within the xtabond2 command.

I have a problem at hand in which I am using GMM for an industry level analysis of percentage of transitions to entrepreneurship, and weighting the regressions based on the number of workers in each industry. My doubt is if the weighting variable should be included as part of the instruments used in the GMM model? I have looked for answers to this online, without any sucess.
The code I have right now is something along these lines:

xtabond2  Perc_trans_entrep L.Perc_trans_entrep      tenure_median        gender log_nemp_median     secondary_education higher_education    reg_1 reg_2 reg_3 reg_4 year_* [aweight=mean_number_workers_ind], gmm (Perc_trans_entrep L.Perc_trans_entrep   tenure_median L.tenure_median       gender log_nemp_median     secondary_education higher_education     ,  lag(1 5)  collapse) iv(  mean_number_workers_ind     reg_1 reg_2 reg_3 reg_4 ,  ) iv( year_*, eq(level)  )   noc twostep robust small

Does this make sense, or should I avoid including the weighting variable in the gmm instruments? Any help would be much appreciated.

Best regards,
Rui

[David Roodman]

In general, I don't think it makes sense to include the weight variable as an instrument or a regressor for that matter--in xtabond2 or any other estimation program. Unless you have some reason to think it really deserves that role in your model.