stochastic frontier analysis with heteroscedasticity using the sftfe command

[M. Yun]

I am trying to estimate a stochastic frontier analysis with heteroscedasticity using the sftfe command written by Belotti and Ilardi (2018), as follows: sftfe lny lnL lnK T, est(mmsle) dist(exp) usigma (z, nocons). I understand this will estimate 3 component technical efficiency, assuming heteroscedasticity. I have question about the post estimation command.
1) "predict u" and "predict jlms" returns the estimated technical efficiency. What is the unit of this value? is it in percent percent form?
2) The post estimation command "predict alpha" returns exactly the same value as technical efficiency. However, I believe this should return firm specific random effects. Does anybody know why this is, or did I do something wrong?

[Andrew Musau]

1) "predict u" and "predict jlms" returns the estimated technical efficiency. What is the unit of this value? is it in percent percent form?
2) The post estimation command "predict alpha" returns exactly the same value as technical efficiency. However, I believe this should return firm specific random effects. Does anybody know why this is, or did I do something wrong?

-u-, -jlms- and -alpha- are options. In your case, you get all the same predictions because you are specifying no options, so the default option is used. All that you are doing is naming your predictions differently. You need:

predict te_u, u
predict te_jlms, jlms
predict fe, alpha

[M. Yun]

Thank you, that really helps!

[M. Yun]

I tried to estimate a heteroscediastic dynamic stochastic frontier production function using the sftfe command, where pde is specified for the model and with the option dynamic as follows:
"sftfe lny lnL lnK t, estimator(pde) d(tn) emean(z1 z2) usigma(z1 z2) vsigma(z1 z2) dynamic rescale". Even with the rescale option, after 175 iterations, I got the following error message.

Hessian is not negative semidefinite
panelsubmatrix(): 3301 subscript invalid
_sftfe_Get_SoL_and_PoSt_ReS(): - function returned error
_sftfe_sf_est_ml(): - function returned error
<istmt>: - function returned error
r(3301);

Can anything be done at this stage to achieve convergence?