A Laplace Stochastic Frontier Model

We propose a Laplace stochastic frontier model as an alternative to the traditional model with normal errors. An interesting feature of the Laplace model is that the distribution of inefficiency conditional on the composed error is constant for positive values of the composed error, but varies for negative values. Therefore, it may be ideally suited for analyzing industries with many forms on or close to the efficient frontier. A simulation study suggests that the model performs well relative to the normal-exponential model when the two-sided error is misspecified. A brief application to US Airlines is provided.