Stochastic Frontier Analysis

[Juan Price Elton]

Hello: I am estimating a SFA with an input distance function and cross section data. I use both a Translog and a Cobb-Douglas production function and different distributions for the inefficiency term. My problem is that when I normalise the variables by the geometric mean both models (based on the TL and the CD functions) work but when I don't normalise by the geometric mean only the model based on a Cobb-Douglas works but the one based on a TL gives very strange results (coefficients with the opposite signs w/r to the expected ones and almost all of them statistically non-significant). Can any one help me with this???? Many thanks , Juan