Work

Nonconvex penalized matrix completion methods for causal inference in panel data

Year: 2023
Type: dissertation
Abstract: potential control outcomes, which improves upon existing methods that rely on the nuclear norm penalty.Remarkably, this upper bound matches the one obtained by the oracle estimator, under an additional condition on the magnitudes of true singular values.Furthermore, we establish the asymptotic normality of the corresponding estimator for the treatment effect, which exhibits a smaller asymptotic variance compared to an existing method.We perform numerical studies to assess the recovery of the potential control outcomes and the estimation of the average treatment effect.Simulations validate our theoretical results, and experiments using real data further demonstrate the promising performance of our proposed method.
Author: 김보영
Language: English
Open Access status: green
OpenAlex ID: W7058469696
Cites:
Cited by:
Related to: 10