How to show the first step of ivpoisson estimation?

[Ahmad Hussein]

Greetings all,

I wonder if it is possible to get the estimation of the first step conducted by the ivpoisson command. If so, what is the command?

Specifically, when using the command: ivpoisson gmm y1 x1...xn (y2 = IV), where y2 is endogenous, Stata reports only the second step estimation. How can I tell Stata to report the first step estimation in a separate table?

[Joao Santos Silva]

Dear Ahmad Hussein,

The command estimates the model by GMM, so there is no first stage as such.

Best wishes,

Joao

[Ahmad Hussein]

Many thanks professor Silva.

I am also confused about another thing regarding the ivpoisson command when using the cfunction option. Stata always tells me that the model is exactly identified even when using instruments larger in number than endogenous variables. As a result, the estat overid command does not work with the ivpoisson cfunction.

To be specific, when using the following command:

ivpoisson cfunction y1 x1...xn (y2 = IV1 IV2)

Stata tells me the model is exactly identified (despite the presence of two instruments (IV1 and IV2) for only one endogenous (y2)). So, the test for overidentification by estat overid does work in this case.


Kindly Joao Santos Silva clarify this point for me because I am confused.

[Fei Wang]

In -ivpoisson gmm-, moment conditions are constructed on the zero correlation between instruments and the error term. So the number of moment conditions grows with the number of instruments and overid test is applicable when instruments are more than endogenous variables. In -ivpoisson cfunction-, the residuals from the first stage are plugged into the main model, and one of the moment conditions is the zero correlation between the first-stage residual and the error term of the main model. No matter how many instruments you have, the number of residuals you obtained from the first stage, as well as their related moment conditions, remains unchanged. In that sense, -ivpoisson cfunction- always fits a just-identified model.