Iteration with nlcom and "expression too long" message

[Giorgia Estefani]

Hallo! I am trying to implement the following program to obtain iterated nonlinear combinations of estimated parameters along with their bootstrap standard errors. I need 20 iterations, however after 10 i get the "expression too long" error message. I am aware of the problem (I read related discussions here in statalist) and I see that my routine is all but efficient. But after several trials I did not get the solution. Anybody can kindly help me out? Many thanks!

use data.dta, clear

program define iterpro, rclass
reg y x z
local b0=1
local b1 "(_b[x]*`b0' + _b[z]*`b0')"
local b2 "(_b[x]*`b1' + _b[z]*`b0')"
local b3 "(_b[x]*`b2'+_b[z]*`b1')"
local b4 "(_b[x]*`b3'+_b[z]*`b2')"
local b5 "(_b[x]*`b4'+_b[z]*`b3')"
local b6 "(_b[x]*`b5'+_b[z]*`b4')"
local b7 "(_b[x]*`b6'+_b[z]*`b5')"
local b8 "(_b[x]*`b7'+_b[z]*`b6')"
local b9 "(_b[x]*`b8'+_b[z]*`b7')"
local b10 "(_b[x]*`b9'+_b[z]*`b8')"

nlcom (a1: _b[x]*`b10') (a2:_b[z]*`b9'), post
nlcom (b11: _b[a1] + _b[a2])

tempname b
matrix `b' = r(b)
return scalar beta = `b'[1,1]
exit
end

bootstrap r(beta), reps(10) seed(1234): iterpro
estat bootstrap

[Andrew Musau]

You need to present a data example that replicates the error to increase your chances of getting a helpful response.

[Giorgia Estefani]

Thanks for the suggestion. The following routine with the attached dataset gives the error, for instance. Whereas if you stop at "local b10" as in the first message it works.

use data.dta, clear

program define iterpro, rclass

reg y x z, r
local b0=1
local b1 "(_b[x]*`b0' + _b[z]*`b0')"
local b2 "(_b[x]*`b1' + _b[z]*`b0')"
local b3 "(_b[x]*`b2'+_b[z]*`b1')"
local b4 "(_b[x]*`b3'+_b[z]*`b2')"
local b5 "(_b[x]*`b4'+_b[z]*`b3')"
local b6 "(_b[x]*`b5'+_b[z]*`b4')"
local b7 "(_b[x]*`b6'+_b[z]*`b5')"
local b8 "(_b[x]*`b7'+_b[z]*`b6')"
local b9 "(_b[x]*`b8'+_b[z]*`b7')"
local b10 "(_b[x]*`b9'+_b[z]*`b8')"
local b11 "(_b[x]*`b10'+_b[z]*`b9')"
nlcom (a1: _b[x]*`b11') (a2:_b[z]*`b10'), post
nlcom (b12: _b[a1] + _b[a2])

tempname b
matrix `b' = r(b)
return scalar beta = `b'[1,1]
exit
end

bootstrap r(beta), reps(10) seed(1234): iterpro
estat bootstrap
program drop iterpro

[Andrew Musau]

I am thinking that you need to drop the program and redefine it every time you run it. The highlighted below is a common line for programs that are run interactively.

clear all
use "data.dta"

capture program drop iterpro
program define iterpro, rclass
reg y x z
local b0=1
local b1 "(_b[x]*`b0' + _b[z]*`b0')"
local b2 "(_b[x]*`b1' + _b[z]*`b0')"
local b3 "(_b[x]*`b2'+_b[z]*`b1')"
local b4 "(_b[x]*`b3'+_b[z]*`b2')"
local b5 "(_b[x]*`b4'+_b[z]*`b3')"
local b6 "(_b[x]*`b5'+_b[z]*`b4')"
local b7 "(_b[x]*`b6'+_b[z]*`b5')"
local b8 "(_b[x]*`b7'+_b[z]*`b6')"
local b9 "(_b[x]*`b8'+_b[z]*`b7')"
local b10 "(_b[x]*`b9'+_b[z]*`b8')"

nlcom (a1: _b[x]*`b10') (a2:_b[z]*`b9'), post
nlcom (b11: _b[a1] + _b[a2])

tempname b
matrix `b' = r(b)
return scalar beta = `b'[1,1]
exit
end

bootstrap r(beta), reps(20) seed(1234): iterpro
estat bootstrap

Res.:

. bootstrap r(beta), reps(20) seed(1234): iterpro
(running iterpro on estimation sample)

Bootstrap replications (20)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5
....................

Bootstrap results                               Number of obs     =        222
                                                Replications      =         20

      command:  iterpro
        _bs_1:  r(beta)

------------------------------------------------------------------------------
             |   Observed   Bootstrap                         Normal-based
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       _bs_1 |   .0062896   .0120174     0.52   0.601    -.0172641    .0298432
------------------------------------------------------------------------------

. estat bootstrap

Bootstrap results                               Number of obs     =        222
                                                Replications      =         20

      command:  iterpro
        _bs_1:  r(beta)

------------------------------------------------------------------------------
             |    Observed               Bootstrap
             |       Coef.       Bias    Std. Err.  [95% Conf. Interval]
-------------+----------------------------------------------------------------
       _bs_1 |   .00628956   .0018723   .01201739    .0012225    .047862  (BC)
------------------------------------------------------------------------------
(BC)   bias-corrected confidence interval

.
end of do-file

[Giorgia Estefani]

Many thanks Andrew, yes I did delete the program at the and of the file. But, you see, if you use one more iteration, as I write below, you get the error message. I think I need to split the nonlinear combinations somehow...

clear all
use "data.dta"

program define iterpro, rclass
reg y x z
local b0=1
local b1 "(_b[x]*`b0' + _b[z]*`b0')"
local b2 "(_b[x]*`b1' + _b[z]*`b0')"
local b3 "(_b[x]*`b2'+_b[z]*`b1')"
local b4 "(_b[x]*`b3'+_b[z]*`b2')"
local b5 "(_b[x]*`b4'+_b[z]*`b3')"
local b6 "(_b[x]*`b5'+_b[z]*`b4')"
local b7 "(_b[x]*`b6'+_b[z]*`b5')"
local b8 "(_b[x]*`b7'+_b[z]*`b6')"
local b9 "(_b[x]*`b8'+_b[z]*`b7')"
local b10 "(_b[x]*`b9'+_b[z]*`b8')"
local b11 "(_b[x]*`b10'+_b[z]*`b9')"
nlcom (a1: _b[x]*`b11') (a2:_b[z]*`b10'), post
nlcom (b12: _b[a1] + _b[a2])

tempname b
matrix `b' = r(b)
return scalar beta = `b'[1,1]
exit
end

bootstrap r(beta), reps(20) seed(1234): iterpro
estat bootstrap

program drop iterpro

[Andrew Musau]

I need to understand what the program is doing, because part of the expression you need to evaluate, stored in the local b11 is below and is indeed too long. Is this what you expect?

use "data.dta", clear
reg y x z
local b0=1
local b1 "(_b[x]*`b0' + _b[z]*`b0')"
local b2 "(_b[x]*`b1' + _b[z]*`b0')"
local b3 "(_b[x]*`b2'+_b[z]*`b1')"
local b4 "(_b[x]*`b3'+_b[z]*`b2')"
local b5 "(_b[x]*`b4'+_b[z]*`b3')"
local b6 "(_b[x]*`b5'+_b[z]*`b4')"
local b7 "(_b[x]*`b6'+_b[z]*`b5')"
local b8 "(_b[x]*`b7'+_b[z]*`b6')"
local b9 "(_b[x]*`b8'+_b[z]*`b7')"
local b10 "(_b[x]*`b9'+_b[z]*`b8')"
local b11 "(_b[x]*`b10'+_b[z]*`b9')"

di "`b11'"

Res.:

 di "`b11'"
(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(
> _b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b
> [x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x
> ]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) +
> _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[
> z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]
> *(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*
> (_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b
> [x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b
> [z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z
> ]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1)))))+_b[z]*(_
> b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) +
> _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b
> [z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z
> ]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]
> *1 + _b[z]*1))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 +
> _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b
> [z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]
> *(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_
> b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1)
> )+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+
> _b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))))+_b[z]*(_b[x]*(_
> b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x
> ]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1
>  + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_
> b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _
> b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)))
> +_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b
> [z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))))))

[Andrew Musau]

So margins allows a longer expression, so you can work with this and modify the program accordingly.

use "data.dta", clear
reg y x z
local b0=1
local b1 "(_b[x]*`b0' + _b[z]*`b0')"
local b2 "(_b[x]*`b1' + _b[z]*`b0')"
local b3 "(_b[x]*`b2'+_b[z]*`b1')"
local b4 "(_b[x]*`b3'+_b[z]*`b2')"
local b5 "(_b[x]*`b4'+_b[z]*`b3')"
local b6 "(_b[x]*`b5'+_b[z]*`b4')"
local b7 "(_b[x]*`b6'+_b[z]*`b5')"
local b8 "(_b[x]*`b7'+_b[z]*`b6')"
local b9 "(_b[x]*`b8'+_b[z]*`b7')"
local b10 "(_b[x]*`b9'+_b[z]*`b8')"
local b11 "(_b[x]*`b10'+_b[z]*`b9')"
*margins, expression((_b[x]*`b11')+ (_b[z]*`b10')) post
margins, expression((_b[x]*`b11'))
margins, expression((_b[z]*`b10'))

Res.:

. margins, expression((_b[x]*`b11'))
Warning: expression() does not contain predict() or xb().
Warning: prediction constant over observations.

Predictive margins                              Number of obs     =        222
Model VCE    : OLS

Expression   : (_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 +
               _b[z]*1) + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 +
               _b[z]*1) + _b[z]*1))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 +
               _b[z]*1) + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 +
               _b[z]*1) + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 +
               _b[z]*1) + _b[z]*1)))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) +
               _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) +
               _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) +
               _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) +
               _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1
               + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 +
               _b[z]*1) + _b[z]*1))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 +
               _b[z]*1) + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))))))

------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       _cons |   .0016015   .0023005     0.70   0.486    -.0029075    .0061104
------------------------------------------------------------------------------

.
. margins, expression((_b[z]*`b10'))
Warning: expression() does not contain predict() or xb().
Warning: prediction constant over observations.

Predictive margins                              Number of obs     =        222
Model VCE    : OLS

Expression   : (_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 +
               _b[z]*1) + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 +
               _b[z]*1) + _b[z]*1))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 +
               _b[z]*1) + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 +
               _b[z]*1) + _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 +
               _b[z]*1)))+_b[z]*(_b[x]*(_b[x]*(_b[x]*(_b[x]*1 + _b[z]*1) + _b[z]*1)+_b[z]*(_b[x]*1 + _b[z]*1))+_b[z]*(_b[x]*(_b[x]*1 +
               _b[z]*1) + _b[z]*1))))))

------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       _cons |   .0024227   .0038077     0.64   0.525    -.0050403    .0098856
------------------------------------------------------------------------------

[Giorgia Estefani]

Many thanks for your time, Andrew! You are really kind. I don't know why but your program does not work with me. I still get the message:

". margins, expression((_b[x]*`b11'))
Warning: expression() does not contain predict() or xb().
expression too long"

For completeness, what I expect is an autoregressive process (with t=20) such that:

b(t)=b(t-1)*beta(1)+b(t-2)*beta(2) with b(0)=1 and b(1)=1, and where beta(1) and beta(2) are the estimated coefficients, on respectively x and z, of the regression y x z.

For this parameter I also need the boostrap standard error.

Many thanks again.

[Giorgia Estefani]

Many thanks for your time, Andrew! You are really kind. I don't know why but your program does not work with me. I still get the message:

". margins, expression((_b[x]*`b11'))
Warning: expression() does not contain predict() or xb().
expression too long"

For completeness, what I expect is an autoregressive process (with t=20) such that:

b(t)=b(t-1)*beta(1)+b(t-2)*beta(2) with b(0)=1 and b(1)=1, and where beta(1) and beta(2) are the estimated coefficients, on respectively x and z, of the regression y x z.

For this parameter I also need the bootstrap standard errors.

Many thanks again.

[Andrew Musau]

I got time to examine your code in #1 and now I see that you do not use the nlcom standard errors. Therefore, just evaluate the expression at the point you define the local, and you can run your bootstrap program without any problems.

clear all
use "data.dta"

program define iterpro, rclass
reg y x z
local b0=1
local b1 "(_b[x]*`b0' + _b[z]*`b0')"
local b2 "(_b[x]*`b1' + _b[z]*`b0')"
local b3 "(_b[x]*`b2'+_b[z]*`b1')"
local b4 "(_b[x]*`b3'+_b[z]*`b2')"
local b5 "(_b[x]*`b4'+_b[z]*`b3')"
local b6 "(_b[x]*`b5'+_b[z]*`b4')"
local b7 "(_b[x]*`b6'+_b[z]*`b5')"
local b8 "(_b[x]*`b7'+_b[z]*`b6')"
local b9 "(_b[x]*`b8'+_b[z]*`b7')"
local b10 "`=(_b[x]*`b9'+_b[z]*`b8')'"
local b11 "`=(_b[x]*`b10'+_b[z]*`b9')'"
nlcom (a1: _b[x]*`b11') (a2:_b[z]*`b10'), post
nlcom (b12: _b[a1] + _b[a2])

tempname b
matrix `b' = r(b)
return scalar beta = `b'[1,1]
exit
end

bootstrap r(beta), reps(20) seed(1234): iterpro
estat bootstrap

Res.:

. bootstrap r(beta), reps(20) seed(1234): iterpro
(running iterpro on estimation sample)

Bootstrap replications (20)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5
....................

Bootstrap results                               Number of obs     =        222
                                                Replications      =         20

      command:  iterpro
        _bs_1:  r(beta)

------------------------------------------------------------------------------
             |   Observed   Bootstrap                         Normal-based
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
       _bs_1 |   .0040242   .0091904     0.44   0.661    -.0139887     .022037
------------------------------------------------------------------------------

.
. estat bootstrap

Bootstrap results                               Number of obs     =        222
                                                Replications      =         20

      command:  iterpro
        _bs_1:  r(beta)

------------------------------------------------------------------------------
             |    Observed               Bootstrap
             |       Coef.       Bias    Std. Err.  [95% Conf. Interval]
-------------+----------------------------------------------------------------
       _bs_1 |   .00402416   .0016969   .00919041    .0006708   .0367184  (BC)
------------------------------------------------------------------------------
(BC)   bias-corrected confidence interval

.

[Giorgia Estefani]

It works! Thanks Andrew: very generous help from you as always. Have a nice day.