Bayesian estimation of a semiparametric stochastic frontier model with persistent and transient inefficiencies

We propose kernel estimators for the densities of persistent and transient inefficiencies in a generalized panel data stochastic frontier model. This approach relaxes the conventional distributional assumptions imposed on the inefficiencies, thereby increasing robustness. A Bayesian sampling algorithm is developed to estimate all model parameters and unobserved entities, as well as the bandwidths needed to construct the densities of persistent and transient inefficiencies. The advantages of our estimator and sampling algorithm are demonstrated through numerical studies based on pseudo-samples. We find that our estimation approach outperforms conventional Bayesian estimation, providing more accurate estimates of firm-level inefficiencies. We also apply the approach to analyze the cost frontier and related characteristics for a panel of large US commercial banks, and discuss differences relative to the traditional Bayesian framework. Our estimator reveals that the high-density region of transient inefficiency varies over time, whereas the conventional approach does not possess such a feature. Moreover, the contribution of persistent inefficiency to the total inefficiency revealed by our approach is slightly more than half of that suggested by the conventional framework.