> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of
> [email protected]
> Sent: 03 February 2004 14:34
> To: [email protected]
> Cc: [email protected]
> Subject: st: Problems Stochastic Frontier Analysis
>
>
> Dear reader,
>
> I have some trouble with performing a Stochastic Frontier
> Analysis in Stata vs8.0. I hope you can help me out, the
> following is the problem:
>
> The generated output from my Stata set-up gives comparable
> but different output then my Limdep set-up (with same input):
>
> the coefficients only differ by small percentages, this
> need not be a
> problem
> Stata gives an extreme big t-value, which seems quiet akward
> as Limdep gives inefficiency numbers between 0 and 1, but
> Stata gives
> 'u' and 'te' values with the same value as 'xb' and the
> fitted dependent
> variable (lnTKhat) (with value around 14). Clearl;y this
> is not correct.
> (Limdep: Inefficiency observation 1=.1651, Stata: te obs
> 1 = 14,0888)
>
> Can you please let me know what the problem is for giving
> wrong 'u' and 'te' values?
>
> see the following log
>
> frontier lnTK lnZT, d(e) cost;
> Stoc. frontier normal/exponential model Number of obs =
> 15
> Wald chi2(1) =
> 1.115e+12
> Log likelihood = 8.5130782 Prob > chi2 =
> 0.0000
>
> --------------------------------------------------------------
> ----------------
> lnTK | Coef. Std. Err. z P>|z| [95% Conf.
> Interval]
> -------------+------------------------------------------------
> ----------
> -------------+------
> lnZT | 1.244566 1.18e-06 . 0.000 1.244563
> 1.244568
> _cons | -4.327623 .0000181 . 0.000 -4.327659
> -4.327588
> -------------+------------------------------------------------
> ----------
> -------------+------
> /lnsig2v | -39.73655 977.6138 -0.04 0.968 -1955.824
> 1876.351
> /lnsig2u | -3.135077 .5163978 -6.07 0.000 -4.147198
> -2.122956
> -------------+------------------------------------------------
> ----------
> -------------+------
> sigma_v | 2.35e-09 1.15e-06 0
> .
> sigma_u | .2085579 .0538494 .1257324
> .3459441
> sigma2 | .0434964 .0224614 -.0005272
> .08752
> lambda | 8.87e+07 .0538494 8.87e+07
> 8.87e+07
> --------------------------------------------------------------
> ----------------
> Likelihood-ratio test of sigma_u=0: chibar2(01) = 7.84
> Prob>=chibar2 =
> 0.003
>
> . predict lnTKhat;
> (option xb assumed; fitted values)
> . predict xb;
> (option xb assumed; fitted values)
> . predict u;
> (option xb assumed; fitted values)
> . predict te;
> (option xb assumed; fitted values)
> . list lnTK lnTKhat xb u te;
>
> +------------------------------------------------------+
> | lnTK lnTKhat xb u te |
> |------------------------------------------------------|
> 1. | 14.16382 14.00888 14.00888 14.00888 14.00888 |
> 2. | 14.32893 14.30058 14.30058 14.30058 14.30058 |
> 3. | 16.78383 16.78383 16.78383 16.78383 16.78383 |
> 4. | 13.81136 13.81136 13.81136 13.81136 13.81136 |
> 5. | 13.54609 13.30281 13.30281 13.30281 13.30281 |
> etc
I know nothing about stochastic friontier analysis, but a bit about
-ml-.
I note the "." reported in the "z" column for lnZt and _cons. Also,
I note that estimate of [/lnsig2v]_cons is -39, and hence sig2v of
virtually zero.
I suspect that problems are arising because you are on the boundary of
the parameter space. It may be that LIMDEP handles this sort of case
differently (check how it parameterizes the model). Without checking out
these numerical issues, I would be reluectant to refer to one set of
estimates as 'wrong'. Is there a option within -frontier- that allows
you to estimate the model for the case when sig2v==0?
Stephen
-------------------------------------------------------------
Professor Stephen P. Jenkins <[email protected]>
Institute for Social and Economic Research
University of Essex, Colchester CO4 3SQ, U.K.
Tel: +44 1206 873374. Fax: +44 1206 873151.
http://www.iser.essex.ac.uk
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
