GMM or 2SLS

[Ishaq Ahmed]

What are the condition for using GMM instead of 2SLS?

[Jeff Wooldridge]

Ishaq: This kind of open-ended question is unlikely to receive an entirely helpful response. Is this something for your research or a homework exercise?

As a general comment, using 2SLS with appropriately robust inference is sufficient, and is less of a "black box." GMM is used when you are trying to gain efficiency due to neglected serial correlation and/or heteroskedasticity. It can also be used with multiple equations to efficiency use information across equations.

[Ishaq Ahmed]

Thanks for your prompt response sir. It is my research that involve Endogeneity issue with 2 model equations. I saw your paper titled "Estimating Panel Data Models in the
Presence of Endogeneity and Selection", my understanding was that 2SLS can take care of the endogeneity. However, I became confused because I later learnt that 2SLS take care of cross-section data but not a panel data and my research is a panel data not cross-section. I was informed that to solve a panel data endogeneity problem is to use GMM. How true it is sir? My research is on Auditing and Corporate Governance. Once again thank you sir.

[Jeff Wooldridge]

Actually, there's nothing wrong with using pooled 2SLS on each equation separately, provided each equation is identified. You can make the inference robust to serial correlation and heteroskedasticity. The only problem with 2SLS is inefficiency. But if your standard errors are small enough to provide sufficiently narrow confidence intervals then you might not worry about what are likely to be modest efficiency gains.

[Ishaq Ahmed]

Thank you most sincerely sir.

[Ishaq Ahmed]

Good day sir.
Is it possible to run GMM simultaneously? Or it is 2SLS that can be ran simultaneousely?