Dealing with endogeneity in stochastic frontier models: A comparative assessment of estimators

Endogeneity poses a major challenge for Stochastic Frontier Analysis, as input choices may be endogenous to unobserved components of the error term, resulting in biased efficiency estimates. This paper compares leading estimators that address this issue, including control-function estimator (Kutlu, 2010), Generalized Method of Moments (GMM) (Tran and Tsionas, 2013) and copula (Tran and Tsionas, 2015) approaches, as well as the instrumental variable based maximum likelihood estimator (Karakaplan and Kutlu, 2017a,b; Karakaplan, 2022). Monte Carlo simulations reveal distinct bias–variance trade-offs: likelihood-based estimators provide more precise efficiency scores, while GMM and copula can be advantageous in specific contexts. An empirical application to the Portuguese thermal power subsector (2006-2021) shows that accounting for endogeneity significantly alters coefficients and efficiency distributions. These results demonstrate that estimator choice affects policy-relevant indicators such as efficiency scores and determinants of cost performance. Despite data limitations, the study underscores the importance of treating endogeneity and provides methodological guidance for applied efficiency analysis. • Endogeneity is a persistent challenge in stochastic frontier models. • We provide a systematic comparison of estimators correcting for endogeneity. • Monte Carlo simulations reveal distinct strengths and weaknesses of each method. • Application to Portugal’s thermal power subsector shows estimator choice matters. • Results offer methodological guidance and policy-relevant insights for regulation.