关键词: 数据包络分析；网络结构；效率分解；交叉效率；共享输入

Abstract: In data envelopment analysis, the existing two-stage cross-efficiency evaluation approach not only has not neutrally decomposed sub-stage efficiency, but also can only be applied to the basic two-stage network structure. A neutral cross-efficiency evaluation approach is present for both basic two-stage network structure and two-stage network structure with shared inputs. In this approach, the overall efficiency of self-evaluation is equal to the weighted sum of the sub-stage efficiency; on the premise of maximizing the overall efficiency of self-evaluation, an optimal set of weights is determined for each decision making unit from the perspective of maximizing the efficiency of each sub-stage; then the final efficiency scores of the whole and sub-stages of the decision making unit are calculated by mutual evaluation. Finally, according to the data in literature ［18］ and ［29］, 24 insurance companies decomposed into basic two-stage network structure and 27 banks decomposed into two-stage network structure with shared inputs are taken as examples. By comparing and analyzing the results of the existing approaches, it is verified that the proposed approach is applicable to a variety of complex network structures, and the approach is practical, reasonable and effective. The contribution of this paper is that the improved efficiency evaluation approach makes up for the limitations of the existing two-stage cross-efficiency evaluation approach and can be applied to many practical problems.

Key words: data envelopment analysis；network structure；efficiency decomposition；cross efficiency；shared input