Hi everyone,

I am facing an issue in determining how I should treat endogenous variable specification in 2SLS IV regression, when the mentioned variable is also interacted in the regression under consideration.

Using on STATA the command ivregress 2sls I have to inlclude the variables that will appear in the last second stage regression therefore also the interaction.

I came up with this question because what will happen when running the code, is that in the first stage when predicting the endogenous variable the interaction that will be included in the first stage also, contains the endogenous variable itself in explaining variation of the endogenous variable itself, it seems to me illogical therefore to use ivregress 2sls in this way also because in the second stage regression the interaction will still be constructed with the endogenous variable and not with the estimated values obtained from the first stage in which the instrument is used.

If anyone has any suggestion or paper to see how to treat this kind of issue it would be well appreciated.

Thanks everyone for you contribute.

Tomaso Mastrocola