The measurement of productive eciency through the estimation of production or cost frontiers has received an increasing amount of attention from both academics and practitioners over the last twenty years. This literature was started by Farrell's seminal paper in 1957, which built upon and . Farrell proposed to measure the eciency of a productive unit in terms of the realized deviations from an idealized frontier isoquant . The empirical identiยฎcation of such a benchmark is the main issue of the literature on eciency measurement. Excellent introductions to this body of literature are provided by Fried, Lovell and Schmidt (1993) and Coelli, Rao and Battese (1998).
It is customary to distinguish between parametric and non-parametric methods to frontier estimation. In the former, the frontier is represented through a functional form (e.g. a Cobbยฑ Douglas or a Translog), derived with econometric techniques . In contrast, the latter methods do not posit any explicit functional form for the frontier and construct it from the observed inputยฑoutput ratios using linear programming techniques. Within the parametric methods, it is possible to distinguish between the deterministic and stochastic approaches to frontier estimation. Deterministic techniques do not accommodate for stochastic shocks to production and cost, and therefore deviations from the frontier are entirely attributed to ineciency. This strong assumption has induced econometricians to abandon deterministic frontiers as a useful model for eciency measurement in favour of stochastic frontier models. In this case, each producer faces a production frontier which is randomly placed by the whole collection of stochastic shocks outside his/her control.