Stochastic DEA Models: Estimating Production Frontiers with Composed Error Models Available to Purchase

In this paper we discuss the Stochastic DEA (SDEA) model introduced in Banker (1988). The linear programming model can be considered a nonparametric quantile regression model where the user chooses a priori the percentage of points below the frontier. Rather than imposing a functional form for production, the SDEA model incorporates the celebrated Afriat conditions to enforce a convex production possibilities set. Recent work on the stochastic frontier models shows how additional assumptions can be placed on the SDEA model to allow a composed error model within the SDEA framework. In this paper, we illustrate these models using a simulated data set. We also apply our SDEA models to the Hildreth (1954) data on corn production.