Estimating Malmquist indices and semi-parametric censored regressions with a coherent data-generating process

This paper presents a semi-parametric method for bootstrapping non-stochastic estimates of Malmquist indices and the second-stage regressions with a coherent data-generating process (DGP) for panel data. It emphasizes the importance of using censoring techniques instead of truncation in the DGP. To keep the panel structure, the fixed effects censored least square (CLS) method is preferred in the algorithm. Two smoothing processes in the DGP using fixed effects CLSs and cross-sectional censored least absolute deviations are proposed and compared. This semi-parametric bootstrap process can be used to delete artificial correlation amongst the estimated efficiencies in panel data. Although the exposition gives only output-oriented indices, the method can easily be extended to the input-oriented model. Finally, an empirical case study is carried out on Irish dairy farms.