Likelihood-ratio test for technological differences in two-stage data envelopment analysis for panel data

This study explores the question of adapting a likelihood-ratio test in the two-stage data envelopment analysis (DEA) framework, where DEA estimates are regressed against external factors. We focus on the hypotheses of testing the technological difference across time periods (or groups) and propose two bootstrapping procedures. Our Monte Carlo (MC) simulation shows that the proposed test has a substantially better-estimated size for the case of smoothing the ‘spurious ones’, rather than removing them. MC simulation also confirms the curse of dimensionality when the input and output variables increase in the production model. Finally, the proposed test is demonstrated in an empirical illustration. • Bootstrap procedures are proposed for testing technological differences. • Smoothing the observations on frontiers leads to better test size. • Trimming the observations on frontiers leads to the distortion of test size. • Empirical data is used to illustrate the method.