Comparative Research And Results Combination Of Frontier Analysis

Frontier analysis method is an important method for measuring technical efficiency. This paper deeply compares common models of frontier analysis method from theoretical and results of calculation aspects, and finally combines the different results of different methods into one result. In theoretical aspect, this paper chooses representative models from non-parametric and parametric methods. For non-parametric methods, this paper chooses the most usual models of data envelop analysis: CCR model, BCC model and super efficiency model, analyzes and compares basic idea, model construction, characteristics of the result for each model, and discusses the common questions of using for the three models. For parametric methods, this paper chooses the most usual model from stochastic frontier analysis and takes the half normal model of panel data for example, researches the process of constructing models and solving models in detail. This paper proposes a general model of panel data, which unifies different distribution models into one model and is proved at length. The general model lays a theoretical basis for explaining the impact that technical inefficiency item is for the result. After analyzing common used models of non-parametric and parametric methods in detail, this paper compares stochastic frontier analysis with data envelopment analysis from the basic assumption and structure of models, the interpretation ant treatment of actual output, the way of constructing frontier, robustness of results, related economic information which can be gotten and the way of analyzing the factors of impacting efficiency six aspects. After comparing and researching theoretic aspect, this paper takes 60 Chinese firms of electronic computer manufacturing industry for example and calculates their technical efficiency from 2000-2005 by those methods above , and then systemically compares results of various methods, finally combines the results of different methods into one result by combined evaluation. In comparison aspect, there are six compared areas: the descriptive statistics, probability distribution, the correlation of the results, the identity of the best and worst decision-making units, the cycle stability of the calculation results and the robustness of the results. In combination aspect, numerical value and rank are both combined. The combined result shows that the combined evaluation has better rank correlation with the original results. Football method which is proposed by this paper improves the main disadvantage of Copeland method. This method not only has very good consistency with the conclusion of each original method, but also its degree of division is better and the conclusion of this method is more accurate. Further more combined evaluation can be widely used for comprehensive evaluation to solve the inconsistent problem of comprehensive evaluation, so it is very worth widely using.