A computational framework for accelerating DEA

We introduce a new computational framework for DEA that reduces computation times and increases flexibility in applications over multiple models and o rientations.The process is based on the identification of frames-minimal su bsets of the data needed to describe the models in the problems-for each of the four standard production possibility sets. It exploits the fact that t he frames of the models are closely interrelated. Access to a frame of a pr oduction possibility set permits a complete analysis in a second phase for the corresponding model either oriented or orientation-free. This second ph ase proceeds quickly especially if the frame is a small subset of the data points. Besides accelerating computations, the new framework imparts greate r flexibility to the analysis by not committing the analyst to a model or o rientation when performing the bulk of the calculations. Computational test ing validates the results and reveals that, with a minimum additional time over what is required for a full DEA study for a given model and specified orientation, one can obtain the analysis for the four models and all orient ations.