Abstract

Abstract There is currently much national and international interest in measuring commercial fishing capacity. Two quantitative methods that will likely be used for this purpose are data envelopment analysis (DEA) and stochastic frontier (SF) production functions. Although both methods can be used to estimate a production frontier, their underlying assumptions and method of solving for the frontier are quite different. One substantial difference is how each model handles noisy data. An understanding of the implications of this difference is important because random variation is likely to exist in commercial fishery catch data. This research uses Monte Carlo simulations to investigate possible finite sample biases attributable to this type of noise when estimating fishing capacity. The results suggest that the mean bias associated with noisy data is often substantially larger for DEA than SF. However, the frequency distributions of the biases from each method show a wide variation in some cases.

Citations