The GMM model

[ALKEBSEE RADWAN]

Hello everyone,

I have a question that I need to be sure about
Can we use the GMM model when the dependent variable is a dummy ?

[Jeff Wooldridge]

GMM is not a model; it is an estimation method. What is the model you are trying to estimate by GMM? Do you want to estimate a dynamic linear probability model using, say, the Arellano and Bond GMM approach?

[ALKEBSEE RADWAN]

GMM is not a model; it is an estimation method. What is the model you are trying to estimate by GMM? Do you want to estimate a dynamic linear probability model using, say, the Arellano and Bond GMM approach?

Yes, I want to estimate the Arellano and Bond GMM approach.
My baseline regression is the Logistic.
Someone tells me that we can't apply the Arellano and Bond GMM approach when the dependent variable is a dummy.
Is this right?
Thank you for your response.

[Sebastian Kripfganz]

As Jeff said, you can estimate a dynamic linear probability model with a binary dependent variable. In principle, one might expect the Arellano-Bond GMM approach to be applicable irrespective of the nature of the dependent variable. I have not really looked into this issue myself, but the following paper suggests that the Arellano-Bond estimator might actually not be consistent in the case of a dynamic linear probability model:
Pua (2015). On IV estimation of a dynamic linear probability model with fixed effects. UvA-Econometrics Working Papers 15-01.

[ALKEBSEE RADWAN]

Jeff Wooldridge
Actually I need to know whether it is possible or not that we use the GMM approach with a dummy-dependent variable (logit regression)?

Thank you

[Jeff Wooldridge]

It’s very important to know whether you want to estimate a dynamic linear probability model or a dynamic logit (or probit) model. The former can be done with GMM. For logit or probit, MLE is currently the best estimation
approach.

[ALKEBSEE RADWAN]

It’s very important to know whether you want to estimate a dynamic linear probability model or a dynamic logit (or probit) model. The former can be done with GMM. For logit or probit, MLE is currently the best estimation
approach.

Thank you so much
Can you provide me the command of MLE estimates?

[Jeff Wooldridge]

I'll look for some slides ...