FINITE SAMPLE EVIDENCE ON THE PERFORMANCE OF STOCHASTIC FRONTIER AND DATA ENVELOPMENT MODELS IN THE ESTIMATION OF FIRM-SPECIFIC TECHNICAL EFFICIENCY USING PANEL DATA
Doctor of Philosophy
In recent years a number of alternative methods have been proposed with which to measure technical efficiency. But we know little of their comparative performance in efficiency measurement. In this study we examine the relative strengths of two different methods--stochastic frontier models using three flexible forms and Data Envelopment Analysis (DEA) in the estimation of firm-specific technical efficiency. To address the limitations of previous studies we utilize Monte Carlo techniques which allow us to control the structure of an underlying technology and two stochastic components. Most stochastic frontier models have focused on estimating average technical efficiency across all firms. The failure to estimate firm-specific technical efficiency has been regarded as a major limitation of previous stochastic frontier models. To overcome this limitation we estimate firm-specific technical efficiency using panel data. We also examine the performance of stochastic frontier models using panel data for three estimators--MLE, GLS and the Within estimator. Our results indicate that for simple underlying technologies the relative superiority of the stochastic frontier models to DEA depends on the choice of functional forms. That is, if the employed form is close to the given underlying technology, it is obvious that stochastic frontier models dominate DEA. If the misspecification of the employed form is serious, DEA is better than stochastic frontier models. For complex underlying technologies the performance of DEA deteriorates markedly. Furthermore, because of an intrinsic problem of DEA, we may not expect that DEA can be used as an efficiency measure for complex underlying technologies. In the application of stochastic frontier models the Within estimator is similar to the other estimators. The similarity may be regarded as a potential solution for two common problems of previous stochastic frontier models--the uncorrelatedness of input levels and technical inefficiency and the dependence of stochastic frontier models on distributional assumptions made on the technical inefficiency. It is hoped that the results of this study will be of use in selecting the appropriate method with which to measure firm-specific technical efficiency in future applied studies.