Conducting system GMM using xtabond2

[Emad Noureldeen]

Dear Stata users,

I am using the System GMM approach for estimation and I want to know whether the code is right or not.
This is the basic equation

ITEMs CEODUAL BDSIZE INDBD SUPSIZE INDSUP SIZE AREC_INVT LOSS AUDFEE GENDER ROA BIG4 AUDOPIN CURR LEV TENURE SWITCH ChangeCPA i.Year i.IND , robust cluster(Stkcd)

Basically, I try to test the impact of corporate governance on disclosure items. The dependent variable is ITEMs and I have 5 independent variables which are CEODUAL BDSIZE INDBD SUPSIZE INDSUP. I also include several control variables including SIZE AREC_INVT LOSS AUDFEE GENDER ROA BIG4 AUDOPIN CURR LEV TENURE SWITCH ChangeCPA. Finally, the year and industry dummies.

I lag the dependent variable and according to the prior studies, they assume that the corporate governance variables are endogenous. So, I write this code and lag them.

xtabond2 ITEMs L.ITEMs CEODUAL BDSIZE INDBD SUPSIZE INDSUP SIZE AREC_INVT LOSS AUDFEE GENDER ROA BIG4 AUDOPIN CURR LEV TENURE SWITCH ChangeCPA i.Year i.IND, gmm(1.ITEMs CEODUAL BDSIZE INDBD SUPSIZE INDSUP) iv(L.CEODUAL L.BDSIZE L.INDBD L.SUPSIZE L.INDSUP SIZE AREC_INVT LOSS AUDFEE GENDER ROA BIG4 AUDOPIN CURR LEV TENURE SWITCH ChangeCPA i.Year i.IND) small twostep robust orthogonal

The questions are:

1- Is the code is right?
2- Should we include all variables that we assume as endogenous in gmm() and all other variables in IV() ?
3- Is the lag above is right or not?

[Sebastian Kripfganz]

In the list of instruments, you probably want to have L.ITEMs, not 1.ITEMs.

For most variables, including them in the iv() option may not be a good idea, even if they are strictly exogenous with respect to the idiosyncratic error component. These instruments enter the level model untransformed, which means that they must be uncorrelated with the unobserved company-specific effects (similar to a random-effects assumption). This is often not a reasonable assumption.

I recommend to have a look at David Roodman's documentation again about how to specify instruments for strictly exogenous variables. Alternatively, the following presentation might be useful:

• Kripfganz, S. (2019). Generalized method of moments estimation of linear dynamic panel data models. Proceedings of the 2019 London Stata Conference.

[Emad Noureldeen]

In the list of instruments, you probably want to have L.ITEMs, not 1.ITEMs.

For most variables, including them in the iv() option may not be a good idea, even if they are strictly exogenous with respect to the idiosyncratic error component. These instruments enter the level model untransformed, which means that they must be uncorrelated with the unobserved company-specific effects (similar to a random-effects assumption). This is often not a reasonable assumption.

I recommend to have a look at David Roodman's documentation again about how to specify instruments for strictly exogenous variables. Alternatively, the following presentation might be useful:

• Kripfganz, S. (2019). Generalized method of moments estimation of linear dynamic panel data models. Proceedings of the 2019 London Stata Conference.

Thanks very much. For the second code, is it right or wrong? because I get a significant results.