Monte‐Carlo Comparison of Conditional Nonparametric Methods and Traditional Approaches to Include Exogenous Variables

Abstract

The aim of this paper is to compare the performance of the conditional nonparametric approach with several traditional nonparametric methods to incorporate the effect of exogenous or environmental variables into the estimation of efficiency measures. To do this, we conduct a Monte Carlo experiment using a translog production function with one output, two discretionary inputs and two exogenous variables to generate simulated data. According to the values of different accuracy measures calculated to evaluate the performance of each method, the conditional data envelopment analysis clearly outperforms all the traditional alternatives.1 There is also another body of literature dealing with those variables in stochastic frontier analysis such as the traditional model developed by Battese and Coelli (1995) or the recent semi-parametric approaches proposed by Kuosmanen (2011, 2012).

Determinants of bank efficiency: evidence from the Latin American banking industry

“…Recently, several models have been developed to provide an appropriate way of accounting for the effect of such variables in non-parametric production models (B adin et al, 2014). The conditional approach introduced by Cazals et al (2002) and extended by Simar (2005, 2007) and Daraio et al (2015) is one such method proposed in the recent literature to overcome the restrictive condition of separability between the input-output space and the space of the environmental variables implicitly assumed by the two-stage approach (Cordero et al, 2016). If the separability condition holds, the factors have no influence on either the shape or the level of the boundary of the attainable set, and the potential effects of environmental factors on the production process are only through the distribution of the inefficiencies.…”

Environmental factors in frontier estimation – A Monte Carlo analysis

“…It is, however, worth emphasizing that cDEA performs comparably well in a substantial number of cases with very small samples, while performance deteriorates for larger sample sizes. Although the latter conflicts with Cordero et al (2016), this divergence seems reasonable given that the authors model environmental factors to affect the inefficiency distribution, but not the frontier as in our case. The good performance of cDEA for small sample sizes is good news particularly for regulatory benchmarking which in most cases is subject to a limited number of observations due to the market structure of network-based industries.…”

“…With respect to rank correlation, cDEA's performance is, with few exceptions, inferior to both LC-SFA and StoNEZD. We find smaller rank correlations of true and estimated efficiency scores than Cordero et al (2016), which is likely due to the different simulation designs. Further, rank correlations vary especially in high impact and multiple-z scenarios, and, in line with Badunenko et al (2012), we find low rank correlations especially if the noise-signal ratio equals one.…”

“…The study finds that the four-stage model introduced by Fried et al (2002) dominates other approaches. Finally, Cordero et al (2016) compare conditional DEA and conditional FDH (cFDH) with DEA approaches. They find that cDEA performs best with respect to identifying efficient units, while other estimators can compete in terms of rank correlations.…”

Sometimes Your Best Just Ain’t Good Enough: The Worldwide Evidence on Subjective Well-being Efficiency

“…The choice of inputs and background variables when implementing two-stage efficiency analyses is subject to debate (Cordero et al 2016;Ravallion 2005). While we cannot settle this debate, we motivate the choices of input and background factors based on existing studies in the literature.…”