Stochastic frontier models assume "iid" (independence) between error terms (those symmetric as well as those that represent inefficiency) across observations. This way we can relatively easily estimate them using frequentist approach (ML method). SURE model structure implies some specific, non-standard form of variance-covariance matrix. That is, we imply some correlation between error terms of different equations, which are complex in SF models. This is not a problem for Bayesian inference (e.g., assume inefficiency to follow Wishard distribution with a specific a priori scale matrix) but I don't think it is doable in frequentist approach.
Anyone has found another papers estimating SFA with time series besides it one suggest by Hasret Balcioglu? I have found few papers (to be exactly just two) and it seens some not quite supported technically.