convergence issues with nlsur (parameters constrained to zero or tending to infinity)

[Luca Tiberti]

Hi all,

I am running nlsur estimations and I am experiencing some difficulty in convergence. Out of 26 parameters to be estimated, 1 is constrained by Stata to zero (ra0) and another one (ck0) tends to infinity (no standard error is estimated). I am joining the estimation output ( nlsur.pdf ). Have some of you an idea of what could be the reason behind these problems?

Thanks in advance,
Luca

[Clyde Schechter]

The better way to show your output is to paste it into your post and surround it by code delimiters. If you are not familiar with code delimiters, read Forum FAQ #12. Attachments are discouraged. Moreover, they limit the range of people who might respond to you because some of us are unwilling to risk downloading attachments from strangers.

[Joro Kolev]

Further, what you re doing there is a Function evaluator program version of -nlsur-. And you are not showing the function program/file itself, you are just showing the command line that executed the function.

Without knowing the function itself, nobody can guess what goes wrong. Or rather anybody can guess anything, and any guess will be just as good as any other.

[Luca Tiberti]

Thanks for your inputs. Here are the full codes I used:

cap program drop nlsurroth
program nlsurroth
version 10

     
     syntax varlist(min=13 max=13) if , at(name)
     tokenize `varlist'
     args shexcl_a shexcl_k age edu avage_k prom_k nkid part_f urban lnexp  nuclear type resid_lnexp
     marksample touse
    qui {
     tempname ca0 ca1 ca2 ca3 ca4 ca5 ca6 ca7
     scalar `ca0' = `at'[1,1]
    scalar `ca1' = `at'[1,2]
     scalar `ca2' = `at'[1,3]
     scalar `ca3' = `at'[1,4]
     scalar `ca4' = `at'[1,5]
     scalar `ca5' = `at'[1,6]
     scalar `ca6' = `at'[1,7]
     scalar `ca7' = `at'[1,8]
     
     tempname ck0 ck1 ck2 ck3 ck4 ck5 ck6 ck7 ck8
     scalar `ck0' = `at'[1,9]
    scalar `ck1' = `at'[1,10]
     scalar `ck2' = `at'[1,11]
     scalar `ck3' = `at'[1,12]
     scalar `ck4' = `at'[1,13]
     scalar `ck5' = `at'[1,14]
     scalar `ck6' = `at'[1,15]
     scalar `ck7' = `at'[1,16]
     scalar `ck8' = `at'[1,17]

     tempname ra0 ra1 rk
     scalar `ra0' = `at'[1,18]
    scalar `ra1' = `at'[1,19]
     scalar `rk' =  `at'[1,20]

     tempname z0 z1 z2 z3 z4 z5
     scalar `z0' = `at'[1,21]
     scalar `z1' = `at'[1,22]
     scalar `z2' = `at'[1,23]
     scalar `z3' = `at'[1,24]
     scalar `z4' = `at'[1,25]
     scalar `z5' = `at'[1,26]
     
     tempname deflatdem_a deflatexp_a deflatexp_a2
    scalar     `deflatdem_a' = 1 
    gen double `deflatexp_a' = (`lnexp' - `deflatdem_a') if `type'==20
    gen double `deflatexp_a2' = `deflatexp_a'^2 if `type'==20
   
    tempvar shexcl_k20
    gen double `shexcl_k20'= 0 if `type'==20
    
    tempvar shexcl_a20
    gen double `shexcl_a20'= (`ca0' + `ca1'*`nkid' + `ca2'*`age' + `ca3'*`edu' + `ca4'*`urban' + `ca5'*`nuclear' + `ca6'*`deflatexp_a' + `ca7'*`deflatexp_a2') + `ra0'*`resid_lnexp' if `type'==20
    
    cap drop `deflatexp_a' `deflatexp_a2' 
    
        tempvar zk 
         gen double `zk' = `z0' + `z1'*`nkid' + `z2'*`avage_k' + `z3'*`prom_k' + `z4'*`nuclear' + `z5'*`urban' if `type'>20
         tempvar gneta_k
        gen double `gneta_k' = exp(`zk')/(1+exp(`zk')) if `type'>20
        tempvar gneta_a
        gen double `gneta_a' = 1 - `gneta_k' if `type'>20
         
        scalar deflatdem_k = 0
        scalar deflatdem_a = 1
        scalar scale =       1 

         tempvar deflatexp_k deflatexp_a deflatexp_a2
        gen double `deflatexp_k' = (`lnexp' + log(`gneta_k') - deflatdem_k) if `type'>20
        
        gen double `deflatexp_a' = (`lnexp' + log(`gneta_a') - log(scale) - `deflatdem_a') if `type'>20
        
        gen double `deflatexp_a2' = `deflatexp_a'^2 if `type'>20

    tempvar shexcl_k21
    gen double `shexcl_k21'= `gneta_k'*(`ck0' + `ck1'*`nkid' + `ck2'*`age' + `ck3'*`avage_k' + `ck4'*`edu' + `ck5'*`urban' + `ck6'*`nuclear' + `ck7'*`deflatexp_k' + `ck8'*`deflatexp_k'^2) + `rk'*`resid_lnexp' if `type'>20
    tempvar shexcl_a21
    gen double `shexcl_a21'= `gneta_a'*(`ca0' + `ca1'*`nkid' + `ca2'*`age' + `ca3'*`edu' + `ca4'*`urban' + `ca5'*`nuclear' + `ca6'*`deflatexp_a' + `ca7'*`deflatexp_a2') + `ra1'*`resid_lnexp' if `type'>20
    drop `deflatexp_k' `deflatexp_a' `deflatexp_a2'

    qui replace `shexcl_a20' = 0 if `shexcl_a20'==.
    qui replace `shexcl_a21' = 0 if `shexcl_a21'==.
    qui replace `shexcl_k20' = 0 if `shexcl_k20'==.
    qui replace `shexcl_k21' = 0 if `shexcl_k21'==.
    
    qui replace `shexcl_a' = `shexcl_a20'+`shexcl_a21'
    qui replace `shexcl_k' = `shexcl_k20'+`shexcl_k21'
   
    }

 end

 #delimit;
nlsur roth @ shexcl_k shexcl_a age edu avage_k prom_k nkid part_f urban lnexp nuclear type resid_lnexp , 
parameters(
ca0 ca1 ca2 ca3 ca4 ca5 ca6 ca7  
ck0 ck1 ck2 ck3 ck4 ck5 ck6 ck7 ck8 
ra0 ra1 
rk 
z0 z1 z2 z3 z4 z5
) 
neq(2) ifgnls;

And here are the results (I avoid posting the initial - many - iteration lines, and start from the last ones):

Iteration 450:  Residual SS =  1.551043
Calculating FGNLS estimates...
Iteration 0:  Scaled RSS =  1400.243
Iteration 1:  Scaled RSS =  1397.388
Iteration 2:  Scaled RSS =  1394.933
Iteration 3:  Scaled RSS =  1393.551
Iteration 4:  Scaled RSS =  1393.541
Iteration 5:  Scaled RSS =  1393.541
Iteration 6:  Scaled RSS =  1393.541
Iteration 7:  Scaled RSS =  1393.541
FGNLS iteration 2...
Iteration 0:  Scaled RSS =  1403.974
Iteration 1:  Scaled RSS =   1403.97
Iteration 2:  Scaled RSS =   1403.97
Iteration 3:  Scaled RSS =   1403.97
Parameter change         =  4.616e-02
Covariance matrix change =  2.004e-06
FGNLS iteration 3...
Iteration 0:  Scaled RSS =      1404
Iteration 1:  Scaled RSS =      1404
Iteration 2:  Scaled RSS =      1404
Parameter change         =  2.438e-03
Covariance matrix change =  1.093e-07
FGNLS iteration 4...
Iteration 0:  Scaled RSS =      1404
Parameter change         =  1.378e-11
Covariance matrix change =  1.643e-16

FGNLS regression 
-----------------------------------------------------------------------
       Equation |        Obs   Parms       RMSE      R-sq     Constant
----------------+------------------------------------------------------
 1     shexcl_k |        702       .   .0224392    0.7071*      (none)
 2     shexcl_a |        702       .   .0415583    0.7917*      (none)
-----------------------------------------------------------------------
* Uncentered R-sq

------------------------------------------------------------------------------
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
        /ca0 |  -.3051822   .2398337    -1.27   0.203    -.7752475    .1648831
        /ca1 |    .002509   .0014185     1.77   0.077    -.0002711    .0052892
        /ca2 |   .0077702   .0018975     4.09   0.000     .0040512    .0114892
        /ca3 |   .0031885   .0020149     1.58   0.114    -.0007606    .0071375
        /ca4 |   .0023744    .001951     1.22   0.224    -.0014495    .0061983
        /ca5 |   .0025963   .0018702     1.39   0.165    -.0010692    .0062618
        /ca6 |   .0499677   .0500723     1.00   0.318    -.0481723    .1481076
        /ca7 |  -.0016288   .0026128    -0.62   0.533    -.0067497    .0034921
        /ck0 |   24.23925          .        .       .            .           .
        /ck1 |   .4162623   1.200133     0.35   0.729    -1.935955    2.768479
        /ck2 |   .1877039    .572713     0.33   0.743    -.9347929    1.310201
        /ck3 |   -.525344   .5650662    -0.93   0.353    -1.632853    .5821655
        /ck4 |    1.56704    1.59704     0.98   0.326    -1.563102    4.697181
        /ck5 |  -1.710966   2.475146    -0.69   0.489    -6.562164    3.140232
        /ck6 |  -6.320843   7.214809    -0.88   0.381    -20.46161    7.819924
        /ck7 |  -7.029256   1.561098    -4.50   0.000    -10.08895   -3.969561
        /ck8 |   1.012668   .5090902     1.99   0.047     .0148694    2.010466
        /ra0 |          0  (constrained)
        /ra1 |  -.0232098   .0042488    -5.46   0.000    -.0315374   -.0148823
         /rk |  -.0437428   .0077543    -5.64   0.000     -.058941   -.0285445
         /z0 |  -4.591388   1.031125    -4.45   0.000    -6.612356   -2.570419
         /z1 |  -.3063783   .0739427    -4.14   0.000    -.4513033   -.1614532
         /z2 |  -.0013501   .0103837    -0.13   0.897    -.0217017    .0190015
         /z3 |  -.0528968   .0287243    -1.84   0.066    -.1091954    .0034019
         /z4 |   .4749318   .1175396     4.04   0.000     .2445583    .7053053
         /z5 |   .1823993   .0908538     2.01   0.045     .0043291    .3604695
------------------------------------------------------------------------------

Thanks again,
Luca