A Bayesian non-parametric stochastic frontier model

Abstract

In this paper, we introduce a new Bayesian non-parametric stochastic frontier (SF) model that addresses the endogeneity problem and relaxes problematic assumptions regarding functional form, and distributional properties. The model can be seen as a competitor to DEA. We show how the model outperforms its parametric counterpart in all critical diagnostic tests. The application we use covers a unique sample of US hotels that operate within competitive clusters. We utilize the efficiency results obtained from this model to shed light on the extent to which performance spillover (i.e. agglomeration effects) may differ based on the varied characteristics of hotels within these clusters. We obtain interesting findings and discuss their implications for hotels contemplating future co-location strategies.