Help with "initial values not feasible" error in frontier command

[Shivam Tawari]

I’m trying to run a code from a paper as part of my analysis, following the steps in a research paper. First, I estimate a model to get the firm effects and errors. Then, in Step 2, I use the frontier command to model the inefficiency on the compustat data.

gen sales_stock = ln(sale)
gen sga_stock = ln(xsga)
gen goodwill_stock = ln(gdwl)  

* Changes made here:
rename fyear year
destring gvkey, replace force
duplicates drop gvkey year, force

encode sic, gen(_sic)
drop sic
rename _sic sic

* Define the stochastic frontier model using maximum likelihood estimation
xtset gvkey year

* step1 : 
xtreg sales_stock sga_stock goodwill_stock i.sic, re
predict alp , u /* Obtain estimates of alpha( time-invariant firm effect) */
predict esl , e /* Obtain estimates of the composed error */

* step 2: 

// frontier alp , distribution ( hnormal )
frontier alp
predict ineff_pers , u /* Predict persistent inefficiency , E(u|e) */
predict eff_pers , te /* Predict persistent efficiency , E(exp ( -u)|e) */

* step 3: 
// frontier esl , distribution ( hnormal )

frontier esl /* This line creates the error */
predict ineff_trans , u /* Predict transitory inefficiency , E(u|e) */
predict eff_trans , te /* Predict transitory efficiency , E(exp ( -u)|e) */

gen MC=ineff_pers*ineff_trans

When I run this code, I get this error:

. frontier esl

initial values not feasible
r(1400);

To fix this, I tried modifying the command like this:

frontier esl sales_stock sga_stock goodwill_stock, distribution(hnormal)

My questions:

1. Am I using the frontier command correctly in Step 2?
2. Should I include variables like sales_stock, sga_stock, and goodwill_stock when working with esl?

I’m new to Stata, and I just want to make sure I’m doing things the right way. Any help would be really appreciated!

Thanks!

[Carlo Lazzaro]

Shivam:
welcome to this forum.
Have you taken a look at -help maximize-?