SFAt: Stochastic frontier analysis on cross-section and panel data

This package provides a collection of methods for doing SFA (Stochastic frontier analysis) based on traditional statistical techniques.

Its development was inspired by some existing software for SFA (R packages frontier, sfa and Benchmarking, Stata commands frontier, xtfrontier and sfcross/sfpanel).

This package is in beta version. It is mostly functional but the interface and some features are still under development. Therefore it is not ready for regular use yet.

Contributions and suggestions are welcome!

Installation

The package is not on CRAN yet. Install it from its GitHub repo.

library(devtools)
install_github("vh-d/SFAt")

Features

cross-section models

• distributions:
  • normal/half-normal
  • normal/truncated normal
  • normal/exponential
• homoskedasticity and heteroskedasticity for both symmetric and inefficiency terms
• conditional mean (BC, 1995) model

panel data models

• panel data conditional mean (BC, 1995) model can be estimated via cross-section model specification
• CSS, 1990 model

To-do:

• more verbose summary output
• panel data models
  • time-invariant model
  • time decay model
  • fixed effects in conditional mean equations in (Battese-Coelli, 1995) model (can be implemented via cross-section model already)
  • fixed effects in conditional variance equations in (Battese-Coelli, 1995) model (can be implemented via cross-section model already)
• explore optimx package for MLE optimization
• add analytic gradient functions
• analytic hessians
• LR test for two SFA objects
• variance of truncated normal distribution