Econometrics and Machine Learning through the Lens of Neyman Orthogonality
Many statistical estimation problems that arise in data-driven decision making and more generally in causal inference, require the estimation of nuisance quantities that are not the focus of the analyst but are simply aides to identify causal models and optimal treatment policies from observational data (e.g. estimating the existing treatment policy when estimating treatment effects). I will present recent results that invoke and extend the principle of Neyman orthogonality to develop estimation methods, such that the estimation error in the target causal model is robust to errors in the estimation of the nuisance quantities. This enables fast mean squared error rates and in many cases also asymptotically valid confidence intervals, even when the nuisance components are fitted via arbitrary ML-based approaches. Moreover, the framework we present, reduces the causal inference problem to a statistical learning problem, thereby enabling the plethora of recent results in statistical machine learning theory to be used for the estimation of complex causal models. I will discuss applications in the estimation of heterogeneous treatment effects, estimation of heterogeneous treatment effects with instruments, estimation with missing data, estimation in games of incomplete information and offline policy learning.
Vasilis Syrgkanis is a Principal Researcher at Microsoft Research, New England, where he is co-leading the project on Automated Learning and Intelligence for Causation and Economics (ALICE). He received his Ph.D. in Computer Science from Cornell University in 2014, under the supervision of Prof. Eva Tardos and spent two years at Microsoft Research, New York as a postdoctoral researcher. His research addresses problems at the intersection of machine learning, economics and theoretical computer science. His work has received best paper awards at the 2015 ACM Conference on Economics and Computation (EC’15), the 2015 Annual Conference on Neural Information Processing Systems (NeurIPS’15) and the Conference on Learning Theory (COLT’19).