Imposing Constraints in a Nonlinear Least-Squares Estimation Function Evaluator Program with Fixed Effects

[Shyam Hari]

I am using the nl function to estimate a nonlinear function with a large number of fixed effects and have written a function evaluator program for the same. The code is

cap program drop nlmyeval
program nlmyeval
  version 17.0
  syntax varlist(min=621 max=621) [if], at(name)
  
  local E: word 1 of `varlist'
  local hc: word 2 of `varlist'
  local l: word 3 of `varlist'
  local a: word 4 of `varlist'
  local b: word 5 of `varlist'
  local id: word 6 of `varlist'
  local inc: word 7 of `varlist'
  local hsize: word 8 of `varlist'
  local temp: word 9 of `varlist'
  local urban: word 10 of `varlist'
  forval i=1/611 {
   local inext=`i'+10
   local d`i' : word `inext' of `varlist'
  }
      
  tempname bl_c bl_inc bl_hsize bl_temp bl_urban
  tempname ba_c ba_inc ba_hsize ba_temp ba_urban
  tempname bb_c bb_inc bb_hsize bb_temp bb_urban

  scalar `bl_c' = `at'[1,1]
  scalar `bl_inc' = `at'[1,2]
  scalar `bl_hsize' = `at'[1,3]
  scalar `bl_temp' = `at'[1,4]
  scalar `bl_urban' = `at'[1,5]
  scalar `ba_c' = `at'[1,6]
  scalar `ba_inc' = `at'[1,7]
  scalar `ba_hsize' = `at'[1,8]
  scalar `ba_temp' = `at'[1,9]
  scalar `ba_urban' = `at'[1,10]
  scalar `bb_c' = `at'[1,11]
  scalar `bb_inc' = `at'[1,12]
  scalar `bb_hsize' = `at'[1,13]
  scalar `bb_temp' = `at'[1,14]
  scalar `bb_urban' = `at'[1,15]
  
  tempvar fe
  gen `fe' = 0 `if'
  forvalues i = 1(1)611{
        tempname bd`i'
        scalar `bd`i'' = `at'[1,15+`i']
        replace  `fe' = `fe' + `bd`i''*d`i'  `if'
  }
  
  tempvar exp_l exp_a exp_b exp_total
  gen `exp_l' = exp(`bl_c'+`bl_inc'*`inc'+`bl_hsize'*`hsize'+`bl_temp'*`temp'+`bl_urban'*`urban') `if'
  gen `exp_a' = exp(`ba_c'+`ba_inc'*`inc'+`ba_hsize'*`hsize'+`ba_temp'*`temp'+`ba_urban'*`urban') `if'
  gen `exp_b' = exp(`bb_c'+`bb_inc'*`inc'+`bb_hsize'*`hsize'+`bb_temp'*`temp'+`bb_urban'*`urban') `if'
  gen `exp_total' = 1+`exp_l'+`exp_a'+`exp_b' `if'
  
  replace `E' = 1/`exp_total'*`hc' + `exp_l'/`exp_total'*`l' + `exp_a'/`exp_total'*`a' + `exp_b'/`exp_total'*`b'+`fe' `if'
  
end

local np = `n'+5*3
mat M = J(1,`np',0.6)

nl myeval @ E hc-d`n', parameters(bl_c bl_inc bl_hsize bl_temp bl_urban ba_c ba_inc ba_hsize ba_temp ba_urban bb_c bb_inc bb_hsize bb_temp bb_urban `names') initial(M) noconstant log

I wish to impose the following constraints,

 constraint 1 _b[bl_inc] = 0
constraint 2 _b[bl_hsize] = 0
constraint 3 _b[bl_temp] = 0
constraint 4 _b[bl_urban] = 0
constraint 5 _b[ba_inc] = 0
constraint 6 _b[ba_hsize] = 0
constraint 7 _b[ba_temp] = 0
constraint 8 _b[ba_urban] = 0
constraint 9 _b[bb_inc] = 0
constraint 10 _b[bb_hsize] = 0
constraint 11 _b[bb_temp] = 0
constraint 12 _b[bb_urban] = 0

With these constraints the estimation effectively becomes a linear regression. The reason I want to do this is to zero out every coefficient associated with the covariates, get just the constant values, and then use that to inform my estimation. However, specifying

constraints(1/12)

does not seem to work (it generates the 198 error code with the message "nlmyeval returned 198
verify that nlmyeval is a function evaluator program"). Writing out the entire model instead of using the function evaluator program is also not feasible because of the large number of fixed effects. I also tried to rewrite the function evaluator program to remove all the above twelve variables, but that also generated the 480 error code "starting values invalid or some RHS variables have missing values".

Is there any way I can impose these constraints? Or maybe another way I can go about performing this restricted estimation? Any help would be appreciated. Thank you so much!