Re: st: xtfrontier tvd

[Gordon Hughes <G.A.Hughes@ed.ac.uk>]

I think that you should think further about what you are trying to 
do.  Just because the method exists in software does not mean that it 
is sensible or appropriate to use it.

Your original message said that the time invariant (ti) models 
converge satisfactorily.  One possible inference from the difficulty 
that you have in estimating the time varying (tvd) model is that this 
model does not fit the data very well.  This is not unusual - in my 
experience it is difficult to obtain sensible estimates from the tvd 
specification and the symptoms that you report are quite typical of 
what happens.

However, if you are sure that you want to persist, then you should 
start the estimation from the parameter values for the converged ti 
specification because this is a restricted version of the tvd 
specification - not just for the usual coefficients on the variables 
but also for other parameters.  Switching between DFP and BFGS will 
sometimes work, though it takes a lot of steps.

Finally, the term "technical efficiency" is a bit misleading.  Look 
at the manual and you will see that --predict xxx, te-- generates 
values of E{exp(-su[it]) | eps[it]} which is the one-sided efficiency 
residual relative to the frontier.  In other words, I suspect that 
your 10% values means that costs/output are 10% above/below the 
efficient frontier.

Gordon Hughes
g.a.hughes@ed.ac.uk

*
*   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/