Stochastic Frontier Analysis and Efficiency Estimation

Abstract

Theoretically, a production function gives the maximum possible output with a given set of inputs. This is different from its common regression counterpart, which specifies the conditional mean of output. The production function defines a boundary or “frontier”, deviations from which can be interpreted as inefficiency. The econometrics of stochastic frontier analysis (SFA) provides techniques for modelling the frontier concept within a regression framework so that inefficiency can be estimated. Obviously, the notion of a frontier can be extended to other representations of technology. Further, with behavioral assumptions like cost minimization, allocative inefficiency can be distinguished from the technical errors. We discuss ways to make this distinction empirically, but in this chapter we concentrate primarily on the estimation of production frontiers and measures of technical inefficiency relative to them. The literature on SFA is now roughly 30 years old and surveys have appeared periodically (Forsund, Lovell and Schmidt (1980), Schmidt (1985–86), Lovell and Schmidt (1988), Bauer (1990) and Greene (1993)). In addition, the literature has been given a textbook treatment by Kumbhakar and Lovell (2000). Aside from reviewing recent advances in SFA, this chapter differs from the earlier surveys in its focus on the use of panel data and attention to questions of econometric and statistical detail. In general, the frontier specifications we consider are variants of the general panel-data regression model: