Mostly harmless simulations? Using Monte Carlo studies for estimator selection. (29th October 2019)

• Record Type: Journal Article
• Authors: Advani, Arun; Kitagawa, Toru; Słoczyński, Tymon
• Is Part Of: Journal of applied econometrics. Volume 34:Number 6(2019)
• Page Start: 893
• Page End: 910
• Publication Date: 2019-10-29
• Subjects: Econometrics -- Periodicals; 330.015195
• DOI: 10.1002/jae.2724
• Languages: English
• ISSNs: 0883-7252

Abstract

Summary: We consider two recent suggestions for how to perform an empirically motivated Monte Carlo study to help select a treatment effect estimator under unconfoundedness. We show theoretically that neither is likely to be informative except under restrictive conditions that are unlikely to be satisfied in many contexts. To test empirical relevance, we also apply the approaches to a real‐world setting where estimator performance is known. Both approaches are worse than random at selecting estimators that minimize absolute bias. They are better when selecting estimators that minimize mean squared error. However, using a simple bootstrap is at least as good and often better. For now, researchers would be best advised to use a range of estimators and compare estimates for robustness.