Performance of Efficiency Measurement by StoNED Compared to DEA and SFA in Cases of Leontief and By-Production Technologies

Stochastic Non-smooth Envelopment of Data (StoNED) is a semi-parametric and stochastic method of efficiency measurement that combines some of the virtues of the classical counterparts, namely Stochastic Frontier Analysis (SFA) versus Data Envelopment Analysis (DEA). Recently, it has been generalised to multi-output technologies by directional distance and ray production functions. First Monte Carlo simulation studies have shown that StoNED offers a promising alternative to DEA and SFA in scenarios with and without noise. However, these studies have exclusively considered technologies modelled by output-oriented translog production functions (such as Cobb-Douglas functions in particular), which allow factor substitution. Moreover, only one of them analyses StoNED for multi-output technologies (thereby using ray production functions). We present complementary results from a series of Monte Carlo simulations with examples of two-input, two-output technologies that are modelled by Leontief functions or are characterised by by-products. Contrary to previous results, DEA then generates competitive or even the best results in scenarios with few noise. Furthermore, StoNED equals the performance of SFA, provided that both methods are adapted to the orientation of the underlying technology, i.e. input-oriented for the Leontief technology