Re: st: Stochastic Frontier Analysis, time-varying effects cost frontier

[Federico Belotti <f.belotti@gmail.com>]

Dear Alexander,

my comments below 

On May 11, 2013, at 6:23 PM, Alexander Lee wrote:

> Dear Statalist members,
> 
> Having read previous posts on the Stochastic Frontier Analysis, I
> still have questions regarding
> its implementation, particularly the so-called Battese and Coeli
> (1995) random time-varying effects
> model is of interest to me.
> 
> My work includes a panel data on several firms, I attempt to explore
> their cost efficiency,
> change of the efficiency scores with time and the impact of the bank's
> type on efficiency (ownership,
> location, etc.). I do that with the -sfpanel command, realized in his
> paper by Prof. F. Belotti. I do
> not assume heteroscedasticity neither in the inefficiency term nor in
> the error term.
> 
> I have some questions on that and would appreciate any insights:
> 
> 
> 1. When I implement a translog form of the frontier model, the
> iterations won't converge
> 
> (BFGS stepping has contracted, resetting BFGS Hessian)
> 
> 
> I believe that all the data is properly scaled and there is a larger
> number of observations.
> 
> I have also tried to do this with -difficult option.
> 

> 
> What could be a reason for this?

Did you impose linear homogeneity in inputs' prices? It is worth noting that such a flexible functional form could be very difficult to estimate, especially in a cost frontier framework.

> 
> 2. If I could estimate the Stochastic Frontier model, which includes
> total costs as dependant
> variable and input prices and outputs as regressors and obtain the
> efficiency scores, I fail to
> understand how the firm types should be accounted for in this
> one-stage model? Should they simply
> be included in the frontier model as new (dummy) variables? However in
> the original 1995 paper I
> could see that firms' effects are included in a separate Inefficiency
> Model, does that mean that the
> inefficiencies obtained from the frontier should be regressed on firm
> types in a separate exercise?

-sfpanel- allows to estimate the Battese and Coelli (1995) model using the following syntax

sfpanel c y p1 p2, cost model(bc95) emean(x1 x2) 

where the option -emean(x1 x2)- allows to simultaneously estimate the so-called inefficiency effects.
Often, the inclusion of exogenous variables to model the mean of the inefficiency could help the identification of the inefficiency term itself (increasing the convergence rate).

hth 
Federico
> 
> 
> Thank you,
> 
> Best regards
> Alexander Lee
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/

-- 
Federico Belotti, PhD
Research Fellow
Centre for Economics and International Studies
University of Rome Tor Vergata
tel/fax: +39 06 7259 5627
e-mail: federico.belotti@uniroma2.it
web: http://www.econometrics.it


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