Efficiency estimation using probabilistic regression trees with an application to Chilean manufacturing industries

We propose smooth monotone concave probabilistic regression trees for the estimation of efficiency and productivity. In particular we modify these techniques to allow for the use of panel data which are often encountered in practice. Probabilistic regression trees provide smooth approximations and at the same time they exploit the versatility of standard regression trees in generating efficiently partitions of the space of the regressors to approximate the unknown frontier. We showcase the new techniques in a large sample of Chilean manufacturing firms.