Help!!! Convergence error in calculating of productivity by SFA - "Kumbhakar, S.C, 2015"

[Louis Nguyen]

Hi everyone,

I am calculating total factor productivity by the SFA model in using STATA.

When I run the TFP estimation, it gives this warning "convergence not achieved"
Do you know the reasons for these errors? and how can I fix it?

Thank you very much!

. matrix b1 = e(b)

. sfmodel lny, prod dist(h) frontier($xvar_prod) usigmas(t) vsigmas()

. sf_init, frontier(b1) usigmas(0 0) vsigmas(0)

. ml max, difficult gtol(1e-5) nrtol(1e-5)

initial:       log likelihood = -2429.7833
rescale:       log likelihood = -2429.7833
rescale eq:    log likelihood = -2429.7833
Iteration 0:   log likelihood = -2429.7833  (not concave)
Iteration 1:   log likelihood = -1588.3033  (not concave)
.......
Iteration 15997: log likelihood =  12249.385  (not concave)
Iteration 15998: log likelihood =  12249.396  (not concave)
Iteration 15999: log likelihood =  12249.407  (not concave)
Iteration 16000: log likelihood =  12249.418  (not concave)
convergence not achieved

                                                Number of obs     =      1,920
                                                Wald chi2(27)     =   6.49e+09
Log likelihood =  12249.418                     Prob > chi2       =     0.0000

------------------------------------------------------------------------------
         lny |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
frontier     |
        lnx1 |  -.0032105    .001224    -2.62   0.009    -.0056094   -.0008116
        lnx2 |   .0008623   .0003733     2.31   0.021     .0001307    .0015939
        lnx3 |  -.0028025   .0008357    -3.35   0.001    -.0044404   -.0011645
        lnx4 |    1.12896   .0152639    73.96   0.000     1.099044    1.158877
        lnx5 |  -.1275546   .0151461    -8.42   0.000    -.1572404   -.0978689
           t |  -.0060357   .0006953    -8.68   0.000    -.0073984   -.0046729
       lnx12 |   .0000333   .0000956     0.35   0.728    -.0001541    .0002206
       lnx22 |  -.0000165   .0000175    -0.94   0.346    -.0000508    .0000178
       lnx32 |   .0000985   .0000693     1.42   0.155    -.0000373    .0002344
       lnx42 |   .0633521   .0078738     8.05   0.000     .0479198    .0787844
       lnx52 |   .0630039    .007834     8.04   0.000     .0476495    .0783583
          t2 |   .0001387   .0000163     8.52   0.000     .0001068    .0001706
    lnx1lnx2 |   .0000402   .0000403     1.00   0.319    -.0000388    .0001191
    lnx1lnx3 |   .0000817   .0000882     0.93   0.354    -.0000912    .0002547
    lnx1lnx4 |  -.0009159   .0003878    -2.36   0.018    -.0016759   -.0001559
    lnx1lnx5 |   .0009364   .0003786     2.47   0.013     .0001944    .0016784
       lnx1t |   .0000484   .0000177     2.73   0.006     .0000137    .0000832
    lnx2lnx3 |  -.0000165   .0000306    -0.54   0.589    -.0000764    .0000434
    lnx2lnx4 |   .0002069    .000124     1.67   0.095    -.0000361    .0004499
    lnx2lnx5 |  -.0002607   .0001251    -2.08   0.037     -.000506   -.0000154
       lnx2t |  -9.77e-06   5.64e-06    -1.73   0.083    -.0000208    1.29e-06
    lnx3lnx4 |  -.0008239   .0002539    -3.24   0.001    -.0013215   -.0003262
    lnx3lnx5 |   .0007864   .0002525     3.11   0.002     .0002915    .0012814
       lnx3t |   .0000393   .0000118     3.33   0.001     .0000162    .0000625
    lnx4lnx5 |   -.063212    .007843    -8.06   0.000    -.0785839     -.04784
       lnx4t |  -.0028864   .0003574    -8.08   0.000    -.0035868    -.002186
       lnx5t |    .002875   .0003556     8.08   0.000      .002178     .003572
       _cons |   .8317727   .0163096    51.00   0.000     .7998064     .863739
-------------+----------------------------------------------------------------
usigmas      |
           t |  -2.543044   .0902784   -28.17   0.000    -2.719986   -2.366101
       _cons |  -4.848508          .        .       .            .           .
-------------+----------------------------------------------------------------
vsigmas      |
       _cons |  -15.37606   .0874759  -175.77   0.000    -15.54751   -15.20461
------------------------------------------------------------------------------
convergence not achieved
r(430);


. sf_predict, bc(bc2_h) jlms(jlms2_h) marginal

[Phil Bromiley]

There are many entries in the pdf documentation on maximum likelihood estimation. There are several things you can try to get convergence. If they don't work, you might try a simpler model and work up.

[Louis Nguyen]

There are many entries in the pdf documentation on maximum likelihood estimation. There are several things you can try to get convergence. If they don't work, you might try a simpler model and work up.

Thank you very much for your support. I will check it.

[ziqie gny]

hi, i m using sfmodel command for Kumbhakar , Lien and Hardaker(2014) model from practitioner' s guide book.This model is a panel data model and estimated with sfmodel.But at the end of the book sfmodel is explained for cross section models and i m not sure if sfmodel is appropriate option for panel data.How can i find help file or application for sfmodel?

[Andrew Musau]

You implement the estimation procedure by Kumbhakar, Lien and Hardaker (2014) in three steps. The first step estimates a panel fixed effects or random effects model. The second and third steps estimate residual inefficiency and persistent efficiency from the error component obtained in the first step - so you have a panel estimator in the first step.

[ziqie gny]

thank you so much for your help.