On the Consistency of the DEA-based Average Technical Efficiency Bootstrap

This paper shows that the bootstrap algorithm for average technical efficiency by Atkinson and Wilson (1995) should be applied with great care for the Data Envelopment Analysis (DEA) estimator if the production frontier is stochastic. A stochastic frontier implies that the DEA estimator is inconsistent, which in turn leads to inconsistent and potentially highly misleading bootstrap confidence intervals. A Monte Carlo simulation study reveals that the bootstrap confidence interval coverage accuracy goes to zero as the sample size increases, even for small contributions of frontier variance to total frontier and efficiency variance.