Policy Learning With Observational Data

In many areas, practitioners seek to use observational data learn a treatment assignment policy that satisfies application‐specific constraints, such as budget, fairness, simplicity, or other functional form constraints. For example, policies may be restricted take the of decision trees based on limited set easily observable individual characteristics. We propose new approach this problem motivated by theory semiparametrically efficient estimation. Our method can used optimize either binary treatments infinitesimal nudges continuous treatments, and leverage where causal effects are identified using variety strategies, including selection observables instrumental variables. Given doubly robust estimator effect assigning everyone treatment, we develop an algorithm for choosing whom treat, establish strong guarantees asymptotic utilitarian regret resulting policy.