This page is devoted to what I am calling Flexible Productivity Analysis or FPA. Over the past 40 years, a large number of models have been developed for productivity/efficiency measurements adding many features such as panel data, time varying, heterogeneity, endogeniety, system estimation and so on [see e.g. works by Battese, Coelli, Greene, Kumbhakar, Lovell, Schmidt, Sickles, Simar, Tsionas, Wilson among others]. There have also been great efforts to use flexible functional forms for technology or inefficiency distributions using e.g. kernel smoothing techniques or DEA.

FPA refers to a nonparametric approach to modelling and estimation of efficiency/productivity models based on low-ranked penalized splines. I use the word flexible to describe several features of the approach: First, using this approach one can estimate semiparametric version of almost any of the models developed in the past 40 years under a unified framework. Second, one has control over the degree of flexibility for each part of the model [anything from parametric to semiparametric to fully nonparametric]. Third, this combined with the trick of "transformation to normal" can handle even more complex models. Finally, there are some degrees of flexibility in using the estimation method [Penalized Least Square, ML, Bayesian, Variational Bayes].

Here, I will gradually add papers and computer codes related to Flexible Productivity Analysis .