st: Re: Finding initial feasible values for xtfrontier, ti cost version, please help!

["Chong, Qi Lin Andrew" <qchong@middlebury.edu>]

Hello all,

I am running an xtfrontier on commercil banks, with the cost specification on stata. I have found David Drukker's guide to finding the initial value for a frontier eqn v. helpful, but he uses the normal specification (production and not cost) and also the example is not in panel form.

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

QUERY1) My question is how to replicate the above steps and find the appropriate inital values for xtfrontier, ti cost. I'm not sure how it changes the format above, in terms of running a straightforward regression, but using

regress logexpense logsum logpdep logpop logdeppop logsumdep logsumpop,
matrix b0 = e(b), ln(e(rmse)^2) , .1



I am left with one missing variable and get the error message



initial vector: matrix must be dimension 10



QUERY1) I am not sure how the initial regression needs to be modified to take into account the fact that it is

1) panel data

2) a cost version

3) and b/c I use a translog cost function, I have interacted outputs and input prices (full model specified below) which I have just included (perhaps incorrectly), in the straightforward regression equation. Any help or guidance on this would be much appreciated!

The reason I need this is because I do not get initial feasible values when I run the program on my data.

*************************************************************************

QUERY2) I specify the model in case the problem arises from my specification. It'll be very helpful if any mistake is detected below, so any comments here will be much appreciated as well!

I am using a translog function for multiple inputs, which I understand as listing all outputs, listing all input prices, listing all outputs interacted with each other (in pairs), listing all input prices interacted with each other (in pairs), and then all possible pairings between outputs and input prices.

I have limited the outputs to 1 and inputs to 2 (by aggregating the classes b/c I was concerned adding too many variables would prevent convergence to a solution in stata), so right now the command is simply

xtfrontier logexpense logsum logpdep logpop logdeppop logsumdep logsumpop, ti cost difficult from(b0, copy)

logexpense = cost of firm
logsum = total output of firm
where logpdep = price of deposits (an input)
logpop = price of operational input
logdepop = price of dep*price of operational input
logsumdep = logsum*logdep
logsumpop = logsum*logpop

I have put these variables into the xtfrontier, and I am not getting feasible values and I am trying to get to them.\

QUERY3)Once I am able to run the frontier I will probably want to split logoutput into its different output components, since now I have added them all up and I'm not sure of the implications. If so, because of the interactions I will be placing a much larger number of variables into the xtfrontier (>20), and I'm not sure whether that will make it much harder to converge to a solution, so any advice on this would be appreciated as well.

If you have any advice or comments about any part of this long request, please feel free to respond to that part only. Thanks a lot for any help you can give!

Andrew



*
*   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/