Bayesian Robust Data Envelopment Analysis With Heavy-Tailed Priors

Abstract

Data envelopment analysis (DEA) remains one of the most widely used methods for evaluating the efficiency of decision-making units (DMUs). However, it is highly sensitive to outliers, especially in cases involving imbalanced data. Classical Bayesian DEA models typically employ Beta distributions as priors, which are not effective in mitigating the influence of outliers. To enhance robustness, we propose a Bayesian DEA model utilizing heavy-tailed priors, such as the Student-t and Cauchy distributions. These priors reduce the impact of outliers, resulting in more stable efficiency estimates. The superiority of the proposed approach is demonstrated through both simulated data and real-world banking data, showing significant improvements over Bootstrap DEA and conventional Bayesian DEA methods.