Abstract: The last paragraph of literature Wei Quanling et al. ^[1] raises an extra resource allocation problem:suppose there are some extra resources which can be given to all or only a part of DMUs, and if we want the allocation to be most beneficial to the whole system, how should the extra resources be distributed. They also point out that the extra resource allocation problem and the selection of DUMs should depend on not only DMUs' efficiency, but also the returns to scale. An on-demand allocation method is proposed to deal with such problem.Firstly, on the basis of technical efficiency and scale elasticity,DMUs'development curves are established. Development curves are used to reveal the corresponding relationships between input change and output change. Secondly, extra resources are divided into several equal parts, and the allocation goal of each part is to achieve the maximum output increase.The allocation will stop once all extra resources are allocated or a part of extra resources does not result in any output increase. Finally, the extra resources obtained by DMU in each allocation process are accumulated to obtain the final allocation results.At the end of this paper, two examples are given to illustrate our method, in the case of ‘one input and one output’ and ‘two inputs, one output’ respectively.The allocation results show that the method proposed in this paper do not reduce the amount of input resources currently occupied by any DMU. Moreover, it can identify the redundancy of extra resources in scale or in structure, so as to timely avoid unnecessary distribution.

Key words: resource allocation, task allocation, Data Envelopment Analysis (DEA), elasticity, Return To Scale (RTS)