One part is stochastic noise. The second part is a strictly positive type variable. This ap-
proach estimates the parameters of a cost frontier. Intuitively, observed deviations from
the frontier can occur for two reasons. Random estimation errors can place an observation
either above or below the frontier. Agent type can only place an observation above the fron-
tier, since a less effective firm can only perform worse than the best, i.e., have higher than
minimum cost. Generalized method of moments (GMM) techniques are used to identify
the parameters of each of the distributions of the composite error as well as the techno-
logical parameters.2 This methodology has several advantages over earlier approaches: it
easily accommodates robust covariance matrix estimation, allows estimation of a full sys-
tem including expenditure share equations, is consistent with profit-maximizing behavior, is
implementable with cross-sectional data, and is computationally undemanding.
As an illustration, I evaluate alternatives for reducing the cost of the CRP. From
its inception in 1985, there have been concerns that asymmetric information regarding pro-
ducers’ reservation values for idled land could inflate the cost of the program. In an effort
to overcome information problems, the program uses an auction format. Producers submit
bids on the rental value they are willing to accept in order to idle an eligible parcel of land.
Initially, the USDA accepted almost all bids that were below a maximum per acre rental
rate, or bid cap, set at a regional level. Producers soon learned the cap for their region and
bids converged to that level (Smith, 1995).
In practice, the CRP thus functioned more like a per-acre Pigouvian subsidy than
an auction. As a result, farmers with relatively unproductive land received surplus pay-
ments above their reservation values. One strategy for reducing these surplus payments is
to redesign the allocation mechanism itself so it does a better job of minimizing the cost of
attaining its objective, given the information asymmetry. Another is to gather data to re-