AW: st: frontier model estimation (Battese & Coelli 92) + error message "initial values are not feasible"

["Kerstin Oriold" <kerstin.oriold@exes.co.at>]

Hi Scott,
many thanks for your helpful information. Regarding the time dummy issue:
they have an impact on the TE scores - the TE scores without dummies in the
xtfrontier regression are much higher than with time dummies. 

Thanks again, Kerstin

-----Ursprüngliche Nachricht-----
Von: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] Im Auftrag von Scott Merryman
Gesendet: Dienstag, 13. April 2010 02:28
An: statalist@hsphsun2.harvard.edu
Betreff: Re: st: frontier model estimation (Battese & Coelli 92) + error
message "initial values are not feasible"

On Sat, Apr 10, 2010 at 9:29 AM, Kerstin Oriold
<kerstin.oriold@exes.co.at> wrote:
> Dear statalisters,
>
> I have a problem in context with the frontier model estimation based on
> Battese & Colli (1992) when working with a panel data set of 11
observations
> over 12 years. Using the respective command:
>
> xtfrontier logpoints logbed logstaff logdischarges d98 d99 d00 d01 d02 d03
> d04 d05 d06 d07 d08, tvd  i(dmu) t(year)
>
> leads to the error message ?initial values are not feasible?. Can anybody
> tell me how to handle this problem ?

You can specify starting values using the -from()-  option.

For example:
webuse xtfrontier1, clear
qui reg lnwidgets lnmachines lnworkers i.t
matrix b0 = e(b), 0,0, ln(e(rmse)^2) , .1
xtfrontier lnwidgets lnmachines lnworkers i.t, tvd from(b0, copy)

See also:  http://www.stata.com/statalist/archive/2003-05/msg00650.html


> If I choose a simpler estimation let?s say without the time dummies or
> without logbeds:
> xtfrontier logpoints logbed logstaff logdis, tvd  i(dmu) t(year)
> xtfrontier logpoints logstaff logdischarges d98 d99 d00 d01 d02 d03 d04
d05
> d06 d07 d08, tvd  i(dmu) t(year)
>
> It does work, however, I would highly appreciate to find a solution for
the
> ?original model? as the time dummies provide valuable information on the
> annual movements related to technology change and the input variable
> ?logbed?  represents the proxy for capital. If I leave out logstaff or
> logdis (represents discharges) some time dummies drop out and I cannot
> calculate anymore any technical efficiency scores for each observation and
> year by using
>
> predict te, te

I don't follow - the predict command will create an efficiency score
for each observation without time dummies.

Scott

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/