Optimal policy learning using Stata

In the footsteps of the recent literature on empirical welfare maximization (EWM), I present a new Stata command called opl to carry out "optimal policy learning", a statistical procedure to design treatment assignments using a machine learning approach. The opl command focuses on three policy classes: threshold based, linear combination, and fixed-depth tree. I show a practical example, based on a real policy case—that is, the popular LaLonde training program—where, by stressing the policymaker perspective, I show how to carry out optimal treatment assignment and the potential operative problems that can come up in applying this procedure to real-world case studies. I will discuss, in particular, problems of “angle solutions”. The presentation offers a general protocol to carry out optimal policy assignment using Stata and contributes to stress the policymaker empirical perspective and related issues arising when carrying out optimal policy assignment in practice.