I am performing a Stochastic Frontier Analysis on unbalanced panel data in R with the package frontier (based on Battese & Coelli 95). I wonder how can we check for autocorrelation / endogeneity. The residuals() function returns the residuals, which consist of both the noise term and the inefficiency term, i.e. residual = y - f(x) = v - u. I am not sure if one can use these residuals to test for heteroscedasticity (Breush-Pagan test) or autocorrelation (LM test). Have you seen this in the literature?
BC 95 was developed long ago so I did not find any material on the endogeneity/autocorrlations of this model. However some other models are available which can help to handle this issue. You may see some recent work of:
Christine Amsler
Subal Kumbhakar
For endogeneity issue, you can use the control function approach to estimate and test. For serial correlation, are you trying to check for v or u or both?In any case, it is much harder since you have unbalanced panel and individual v and u are not identified. I have not seen any work on testing for serial correlation in SFM. If you are worry about serial correlation my suggestion is to use complete dynamic modeling on the frontier.
Not sure if this helps but there is a Stata package called xtsfkk that handles endogeneity in panel SFA models. It was only released in 2022, I am using it for my current research and it it a great routine.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression