Boostrapping system of equations

[Prachi Mishra]

Hi, I am trying to bootstrap standard errors for a system of equations (its a 2SLS where I create the first stage fitted values, and use in second stage as regressors; see my code below). But it produces ONLY the second stage output; how do I retrieve the first stage output too with bootstrapped standard errors?

Many thanks

* Produce bootstrap standard errors

capture program drop bootiv
program bootiv, eclass
{
* 1st stage T11 col1
reg inq d_sob past_rel low_score medium_score high_score male_dum log_age ///
i.tier i.qtr_yr, cl(fid) /* 1st stage */
predict xbvar, xb

** 2nd stage col 2 t11
reg default2_360 xbvar past_rel low_score medium_score high_score male_dum log_age ///
log_loan i.new_acctid i.tier i.qtr_yr if e(sample), cl(fid) /* 2nd stage */
sum xbvar if e(sample), detail
drop xbvar
}

end

set seed 1
bootstrap, reps(5) seed(123): bootiv
outreg2 using "BS check.xls", append ctitle("Boostrapped SE") bdec(4) tdec(4) se

[Joro Kolev]

The easiest is just to cut out the code that starts with ** 2nd stage col 2 t11.

[Prachi Mishra]

Thank you. But would it produce the bootstrapped standard errors for the first stage (that corresponds to bootstrapping the two together?);

[Felix Bittmann]

You can save the matrices of both regressions and then combine them in a vector. Here are some ideas:
https://www.stata.com/statalist/arch.../msg01020.html
https://www.stata.com/support/faqs/s...pping-vectors/

I managed to do it like this but it needs some refinements. Hope it helps.

clear all
sysuse auto

cap program drop test1
program define test1, eclass
    tempname reg1 reg2
    reg price mpg
    matrix `reg1' = e(b)
    
    
    reg trunk length foreign
    matrix `reg2' = e(b)
    matrix bb = (`reg1',`reg2')
    matrix colnames bb = "A" "B" "C" "D" "_cons"
    ereturn post bb
    ereturn local cmd="bootstrap"
end



bootstrap, reps(100) seed(123): test1

[Prachi Mishra]

@ Felix .. many thanks. I tried the process you suggested; the matrix is showing different coefficients for 2nd stage than what the regression results show (or what is produced without bootstrap). the coefficients should be identical.