Another approach to data envelopment analysis in noisy environments: DEA+

In this paper a DEA+ labeled approach for efficiency measurement in the stochastic case is presented along with a consistency proof and some preliminary evidence illustrating the small sample performance, DEA+ can basically handle multi-output technologies like standard DEA but allows to filter noise, that might have disturbed production and unlike a related approach does not require panel data. Consistency of DEA+ relies on the assumption of i.i.d. distributed and bounded noise and requires radial efficiency measurement. First Monte Carlo experiments show that a DEA+ based average inefficiency estimator performs well for samples of size n = 100 in one-output, two-input settings compared to the corresponding Stochastic Frontier Estimator. Sensitivity of DEA+ performance with respect to parametrization of noise is weak, but higher noise contribution requires much larger sample size for satisfactory results.