Special Issue on Model Selection and Inference Published: Double/Debiased Machine Learning for Treatment and Structural Parameters

The Econometrics Journal has published a Special Issue on Model Selection and Inference. This Special Issue and the four regular papers published with it, are available free of charge from http://onlinelibrary.wiley.com/doi/10.1111/ectj.2018.21.issue-1/issuetoc.

The paper in this Special Issue arises out of the invited presentations given in The Econometrics Journal Special Session on this topic at the 2016 Annual Conference of the Royal Economic Society. This Special Session was organized by Richard Smith, then Managing Editor of The Econometrics Journal, and chaired by Andrew Chesher, then President of the Royal Economic Society, with Richard Smith also overseeing the editorial process for the submitted paper arising from the Special Session. Chris Hansen (University of Chicago) presented “Model selection and post-model selection inference in economic applications”. Bruce Hansen (University of Wisconsin, Madison) presented ”Shrinkage estimation in vector autoregressions”. Both presentations are available at http://www.res.org.uk/view/Webcasts-Special- Sessions-video-16.html.

The paper ”Double/debiased machine learning for treatment and structural parameters” by Victor Chernozhukov (MIT), Denis Chetverikov (UCLA), Mert Demirer (MIT), Esther Duflo (MIT), Christian Hansen (University of Chicago), Whitney Newey (MIT) and James Robins (Harvard University) was, just before its publication, among the Journal”s most downloaded unpublished articles (http://www.onlinelibrary.wiley.com/doi/10.1111/ectj.12097/abstract). Its main contribution is the provision of general and simple valid procedures for root-N consistent estimation and inference on a low dimensional parameter of interest, typically a causal or treatment effect parameter, in the presence of a high-dimensional or highly complex nuisance parameter. It discusses applications to learning the main regression parameter in a partially linear regression model, the average treatment effect and treatment effect on the treated under unconfoundedness, and the local average treatment effect in an instrumental variables setting.