Statistical inference in efficiency analysis with applications

This thesis is a contribution to frontier analysis and its application to developing areas in Morocco. It focuses on both nonparametric and parametric frontier methods and mainly proposes inference methods for efficiency under weak conditions. In a stochastic frontier analysis, a possible dependence between the two error terms is considered for both cross-sectional and panel data. In order to validate our models, several tests are performed and analyses of the sensitivity to outliers are established. Confidence intervals are constructed using the smoothed bootstrap for the nonparametric approach and two extended parametric percentile bootstrap algorithms for the parametric one. Both are illustrated using simulated and real data. For panel data, a variety of alternative deterministic time effect models are proposed, allowing e.g. for periodicity and disappearing inefficiency over time. Efficiency scores generally suggest the inefficiency of the handled sectors, but results should be interpreted carefully given the unavailability of certain factors.