Bayesian Nonparametric Inference in Bank Business Models with Transient and Persistent Cost Inefficiency

Abstract

This paper introduces a novel econometric framework for identifying and modeling bank business models (BBMs), which dynamically evolve in response to changing financial and economic conditions. Building on the stochastic frontier literature, we extend the traditional cost-efficiency models by decomposing inefficiency into persistent and transient components. We propose a Bayesian nonparametric approach that adapts to the data through an infinite mixture model with predictor-dependent clustering, enabling a flexible classification of banks into distinct business models. Our method, based on the Logit Stick-Breaking Process (LSBP), provides a data-driven way to capture the heterogeneity in bank strategies, allowing for dynamic transitions between business models over time. This model offers a significant advancement over existing parametric and kernel-based approaches by combining the scalability of nonparametric methods with efficient computational routines. We apply the model to a dataset of European banks and identify four distinct business model clusters, providing novel insights into the evolution of bank performance and efficiency. Our findings contribute to the growing literature on the identification and measurement of bank business models, offering valuable implications for policy and regulatory frameworks.