cmp: two-way clustered standard errors

[Gary Yu Gan]

Hello everyone,

I am currently working on estimating a 2SLS model with two endogenous variables (x1 and x2) and two equations in the first stage using the -cmp- command.

First stage equations:

• x1 = z1 + covariates + epsilon1
• x2 = z2 + covariates + epsilon2

Following that, I estimate a second-stage equation to compare the coefficients for x1^hat and x2^hat.

Second stage:

• y=x1^hat + x2^hat + covariates + epsilon3

According to Abadie et al. (2023, QJE), I need to adjust the standard errors by clustering at x1 and x2. However, it seems that the command is unable to calculate two-way cluster-robust standard errors.

Is there a way to estimate two-way clustered standard errors when using the -cmp- command? Alternatively, are there other manual approaches to estimate the multi-equation 2SLS model with two-way clustered standard errors?


Thank you very much for your help!

Gary

[David Roodman]

The waldtest command, part of my boottest package, should be able to do that. Example:

webuse laborsup
cmp setup
cmp (fem_work = other_inc fem_educ kids) (other_inc = fem_educ kids male_educ), ind($cmp_probit $cmp_cont)
waldtest [fem_work]other_inc, cluster(kids fem_educ)