ivregress 2SLS: instrument only with varlist_iv not varlist_1

[Jack Reimer]

Hello I am trying to specify the following regression

reg kids educ age age2 south east west i.time  black ///
    c.age#c.educ south#c.educ east#c.educ west#c.educ black#c.educ

Where `south`, `east`, `west` and `black` are binary variables and `time` is a factor variable ranging between 1 and 7. All variables except `educ` are to be treated as exogenous and I can instrument for `educ` using the following variables `meduc and `feduc`. I do this with the following `ivregress` command

ivregress 2sls kids age age2 south east west i.time black ///
    (educ c.age#c.educ south#c.educ east#c.educ west#c.educ black#c.educ = ///
    meduc feduc ///
    c.age#c.meduc c.age#c.feduc ///
    south#c.meduc south#c.feduc ///
    east#c.meduc east#c.feduc ///
    west#c.meduc west#c.feduc ///
    black#c.meduc black#c.feduc), vce(robust)

Which runs just fine with the following description of endogenous variables and instruments

Instrumented:  educ c.age#c.educ 1.south#c.educ 1.east#c.educ 1.west#c.educ 1.black#c.educ
Instruments:   age age2 south east west 2.time 3.time 4.time 5.time 6.time 7.time black meduc feduc c.age#c.meduc c.age#c.feduc 1.south#c.meduc 1.south#c.feduc 1.east#c.meduc 1.east#c.feduc 1.west#c.meduc 1.west#c.feduc 1.black#c.meduc 1.black#c.feduc

My Question:

Looking at the docs, it seems like I am getting this output because ivregress treats varlist_1 (i.e. `age`-`black`) *and* `varlist_iv` (i.e. everything specified in the parentheses after `=`) as instruments for my endogenous variables.

Can I just instrument for my endogenous variables with varlist_iv?

I am unsure if this is just a difference in nomenclature but I would prefer to have only the 12 instruments listed in the parentheses rather than all of my covariates as well.

Thank you

[Jack Reimer]

Note that the true motivation for this question is that I am receiving different output when I manually specify each of my first stages, save the fitted values, and then run them in a normal `reg` command.

[Sebastian Kripfganz]

Excluding the exogenous regressors in varlist1 from the first-stage regression generally turns the estimator inconsistent, unless those exogenous regressors in varlist1 are uncorrelated with the endogenous regressors in varlist2. See for example Wooldridge (2002), Econometric Analysis of Cross Section and Panel Data, Section 5.1.2.