CMP confusion

[danny klinenberg]

Hello all,

I am running a multivariate Tobit with constrained parameters using cmp. However, I'm having trouble identifying the underlying econometrics. I think I know how it's set up but I would appreciate someone confirming the following is what cmp is doing:

Setup: I have two outcomes, y1 and y2, with one explanatory variable, x. y1 and y2 are censored at zero, so let's pretend there is an unobserved y1* and y2* that are latent uncensored variables. I believe there is a correlation in the errors between the two regressions so I want to estimate:

y1=b0+b1*x+e1
y2=a0+a1*x+e2

e1 and e2 are joint normally distributed and b1=-a1 from theory.

Code: I then run the following lines in STATA:

---

constraint 1 [y1]x = -[y2]x

cmp (y1=c.x) (y2=c.x), indicators("cond(y1>0, $cmp_right, $cmp_cont)" "cond(y2>0, $cmp_right, $cmp_cont)") robust constraint(1)

---

Question: Is my theoretical setup what cmp is doing "under the hood"? I know there are various levels to add the dependency when thinking about multivariate tobits (i.e. the latent variable or observed variable).

Thanks for the help!

[David Roodman]

Yes, your description of the data-generating process matches what cmp assumes. I think the 2nd and 2rd arguments in the cond() expressions are backwards.

[danny klinenberg]

Thank you David! I appreciate the help. Everything *appears* to be running smoothly. Thanks again for making the command and answering my question.