Causal Inference, Neyman Orthogonality And M/P Redundancy, With Application To Productivity And Efficiency

Abstract

Causal inference is about statistical testing for relationships, not about prediction. Machine learning is not so good at this. I will introduce the relatively new area of causal inference which adapts ML methods to the fundamental task of scientific discovery. Central to the methodology is the concept of Neyman orthogonal moment conditions. I will connect this concept with moment and parameter redundancy, a condition introduced by Prokhorov and Schmidt (2009) within General Method of Moments estimation, and I will show how this condition works out for a large class of models of production. The approach allows us to obtain robust post-ML inference of such important causal quantities as returns to scale and factor productivity.

Presenter

Artem Prokhorov is a professor at the Discipline of Business Analytics of the University of Sydney Business School and an internationally recognised scholar in econometrics. Professor Prokhorov's research interests are in the field of theoretical and applied econometrics, with applications in business, finance, risk management, labour and health economics, among other fields, and have intersections with statistical machine learning and high-dimensional statistics. His research has appeared in top field journals such as the Journal of Econometrics, Journal of Financial Econometrics and Journal of Banking and Finance and attracted extensive competitive research funding from Australian Research Council (ARC), Fonds de recherche du Québec – Société et culture (FQRSC), Russian Science Foundation (RNF), Social Sciences and Humanities Research Council of Canada (SSHRC), as well as from numerous industry partners.