Title: "NMM-StoNED: a normal mixture model based stochastic semi-parametric benchmarking method"

DOI: 10.15224/978-1-63248-058-3-95
Page(s): 65 - 69
Authors: XIAOFENG DAI   


This paper presents a novel benchmarking tool, NMM-StoNED, which identifies the best practices closely located with each decision making unit (DMU) in the input-output space. Unlike the conventional techniques such as DEA where the success recepies of the benchmarks may not be transferable to all DMUs given their differences in, e.g., the operational scales, best practices identified by this method do not suffer from these problems and offer more practical values. NMM-StoNED is a specific configuration of the clustering and efficiency estimation algorithms in the benchmarking framework previously presented. This combination is able to cluster DMUs into less ambiguous groups and model the inefficiencies in a stochastic semi-nonparametric framework, which produces more accurate results than conventional benchmarking techniques such as DEA or other combinations such as the integration of K-means and StoNED. The performance comparison between NMM-StoNED and DEA has previously been reported, and the superiorities of StoNED over other productive efficiency analysis methods have been thoroughly investigated. Here we focus on showing the advantages of NMM in the clustering based benchmarking framework, for which, an empirical study using the Finland energy regulation data was conducted. This study contributes in its systematic evaluations on the performance of NMM-StoNED under various conditions which provide solid specifications on this algorithm, availing its practical use.