Two Stage Poisson IV (Panel Data)

[Jorge Vlachopoulos]

Hi there,

I am trying to run the following:

1st stage: reghdfe x1 = z1 c1, absorb(category time) vce(robust)
predict double v2hat1, r
reghdfe x2 = z2 c1, absorb(category time) vce(robust)
predict double v2hat2, r
2nd stage: ppmldhfe y = x1 x2 v2hat1 v2hat2 c1, absorb(category time) vce(robust)

I am not sure how to bootstrap the standard errors correctly (and whether this approach is correct). I am following an earlier post discussing ivpoisson which looks into this problem but I feel that there is no clear answer on the correct wat to bootstrap.

My attempt is the following:

* Define the process as a program to bootstrap
program define mybootstrap, rclass
* First regression with inmodel_intro
reghdfe inmodel_intro inmodel_intro_p price, absorb(id_marca wdate) vce(robust)
predict douncle v2hat1, r

* Second regression with outmodel_intro
reghdfe outmodel_intro outmodel_intro_p price, absorb(id_marca wdate) vce(robust)
predict douncle v2hat12, r

* Running the PPMLHDFE regression
ppmlhdfe quantity inmodel_intro outmodel_intro v2hat1 v2hat2 price, absorb(id_marca wdate) vce(robust)

* Store the coefficients in r() as scalars for bootstrap
return scalar b_inmodel_intro = _b[inmodel_intro]
return scalar b_outmodel_intro = _b[outmodel_intro]
return scalar b_v2hat1 = _b[v2hat1]
return scalar b_v2hat2 = _b[v2hat2]
return scalar b_price = _b[price]
end

* Run the bootstrap, referencing the returned scalars
bootstrap r(b_inmodel_intro) r(b_outmodel_intro) r(b_v2hat1) r(b_v2hat2) r(b_price), reps(500) seed(123): mybootstrap

Thanks!

[Jorge Vlachopoulos]

As I am working on the code. Here is an update that works but very slow.

* Define the process as a program to bootstrap
program define my_bootstrap, rclass

* First regression with inmodel_intro
xtreg inmodel_intro inmodel_intro_p price i.marca i.wdate, fe
predict double v2hat1, r

* Second regression with outmodel_intro
xtreg outmodel_intro outmodel_intro_p price i.marca i.wdate, fe
predict double v2hat2, r

* Running the PPMLHDFE regression
ppmlhdfe quantity inmodel_intro outmodel_intro v2hat1 v2hat2 price, absorb(id_marca wdate)
return scalar b_inmodel_intro = _b[inmodel_intro]
return scalar b_outmodel_intro = _b[outmodel_intro]
return scalar b_v2hat1_1 = _b[v2hat1]
return scalar b_v2hat1_2 = _b[v2hat2]
return scalar b_price = _b[price]

drop v2hat1 v2hat2

end

* Run the bootstrap, referencing the returned scalars
bootstrap r(b_inmodel_intro) r(b_outmodel_intro) r(b_v2hat1_1) ///
r(b_v2hat1_2) r(b_price), ///
reps(100) seed(123) cluster(id_sku) :my_bootstrap