Efficiency measurement for multi-product industries: A comparison of classic and recent techniques based on simulated data

Abstract

In this paper, a cost function is used to generate the data of six samples of firms producing three outputs by means of two factors; this data-generation process is designed to reflect some structural characteristics of multi-product industries: e.g., the production levels of the different outputs are highly correlated and big producers are less frequent than small ones. A known amount of inefficiency and random noise is then added to each production plan. Finally we compare the “true” inefficiency levels to those estimated through the following techniques: stochastic frontiers, DEA, and stochastic DEA (two original models are also developed). All the “classic” techniques (translog cost function, CRS DEA, VRS DEA) perform well, although the first one can not achieve a satisfactory decomposition of efficiency into its allocative and technical causes. The stochastic DEA models can outperform the “classics” in some specific situations, but on average they cannot compete with older techniques; however, the two new stochastic DEA models perform better than the standard one.