We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

econ.EM

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Economics > Econometrics

Title: Causal Inference in Possibly Nonlinear Factor Models

Authors: Yingjie Feng
Abstract: This paper develops a general causal inference method for treatment effects models with noisily measured confounders. The key feature is that a large set of noisy measurements are linked with the underlying latent confounders through an unknown, possibly nonlinear factor structure. The main building block is a local principal subspace approximation procedure that combines $K$-nearest neighbors matching and principal component analysis. Estimators of many causal parameters, including average treatment effects and counterfactual distributions, are constructed based on doubly-robust score functions. Large-sample properties of these estimators are established, which only require relatively mild conditions on the principal subspace approximation. The results are illustrated with an empirical application studying the effect of political connections on stock returns of financial firms, and a Monte Carlo experiment. The main technical and methodological results regarding the general local principal subspace approximation method may be of independent interest.
Subjects: Econometrics (econ.EM); Methodology (stat.ME); Machine Learning (stat.ML)
Cite as: arXiv:2008.13651 [econ.EM]
  (or arXiv:2008.13651v3 [econ.EM] for this version)

Submission history

From: Yingjie Feng [view email]
[v1] Mon, 31 Aug 2020 14:39:36 GMT (151kb,D)
[v2] Fri, 26 Feb 2021 07:20:38 GMT (415kb,D)
[v3] Wed, 13 Oct 2021 13:42:41 GMT (422kb,D)

Link back to: arXiv, form interface, contact.