Abstract
This study develops an innovative Bayesian stochastic frontier model that addresses two critical limitations in efficiency analysis: the curse of dimensionality in high-dimensional production environments and the failure to respect natural groupings among input variables. Our methodology combines group LASSO regularization with adaptive priors informed by data envelopment analysis, creating a unified framework that leverages both parametric and nonparametric efficiency information. The group-structured penalties preserve economically meaningful variable relationships while enabling automatic selection of relevant input clusters. An efficient Markov chain Monte Carlo algorithm facilitates posterior inference under this complex hierarchical specification. Extensive numerical experiments demonstrate remarkable improvements in estimation accuracy, particularly in high-dimensional and sparse settings. Application to commercial banking data reveals the method's ability to identify core operational drivers and uncover bimodal efficiency distributions, suggesting distinct strategic positioning among financial institutions.