Stochastic frontier model sfpanel

[StochasticGrey]

Hi,

Im trying to create a Stochastic frontier model , I ahve gathered my data and put the inputs,outputs and variables in their appropriate places but my iteration wont converge !! Any suggestions ?

do i need to create vectors for my inputs and outputs or are they automatically created ?

[RichardSimper]

We have found that to begin with - starting values are sometimes necessary. run an OLS regression of your variables and copy and paste these values into svfrontier(...). Next start to play with the tech(nr 40 bfgs 270 dfp 120 bhhh 1000 nr 2000) changing the number of iterations in each tech and changing them around. I would love to hear from anyone who has a better way of doing this to achieve convergence - very time consuming. Here is an example reg LTCP1 LP21 LP21LP21 LY1 LY2 LY3 LY1LY1 LY2LY2 LY3LY3 LY1LY2 LY1LY3 LY2LY3 LY1LP21 LY2LP21 LY3LP21 T
// Frontier Estimation
sfpanel LTCP1 LP21 LP21LP21 LY1 LY2 LY3 LY1LY1 LY2LY2 LY3LY3 LY1LY2 LY1LY3 LY2LY3 LY1LP21 LY2LP21 LY3LP21 T, cost m(bc95) svfrontier(0.399 -0.001 -1.176 1.325 3.121 -0.009 -0.468 0.020 0.273 -0.220 0.043 0.111 0.040 -0.230 -0.167 0.017) svemean(-10 1 -10 -1 1) e(CR5 GDPGR RealI ASSGDP) distribution(tnormal) tech(nr 40 bfgs 270 dfp 120 bhhh 1000 nr 2000) iterate(4000) feshow

There has to be a do loop code out there?

[begum guler]

Hi,

Many thanks for the code which proved quite helpful. The convergence finally achieved. But may I ask how you get the starting value of LTCP1? which value do you insert? did you also insert the value of the constant term from OLS?