3

Relaxing Instrument Exclusion with Common Confounders

Finding excluded (and exogenous) instruments is hard. We consider the situation where instruments are excluded only conditional on some unobserved common confounders, for which relevant proxies exist. Using insights from proximal learning, we can identify exogenous variation in the instruments to then identify a causal effect of a treament on an outcome. All our relevance assumptions are testable, while as usual in IV, the assumption of exclusion conditional on unobservables is not (up to specification tests). Importantly, exclusion conditional on unobservables for which proxies exist may be a weaker assumption than exclusion conditional on observables only.

Christian Tien

Relaxing Instrument Exclusion with Common Confounders

Instrumented Common Confounding

Instrumented Common Confounding (ICC) identifies causal effects with instruments, which are exogenous conditional on some unobserved common confounders. Suitable examples of this setting are various identification problems in the social sciences, dynamic panels, and problems with multiple endogenous confounders. The ICC identifying assumptions are closely related to those in mixture models, negative control and IV. We prove point identification with outcome model and alternatively first stage restrictions, and present the causal effect of education on income as a motivating example.

Christian Tien

Instrumented Common Confounding

Found in the (Random) Forest

In this paper written in about two days for the Econometric Game 2021, we use employ Generalised Random Forests to estimate a local average partial effect that comes closest to a causal effect of Airbnb on house prices. The estimated average treatment effects show a nuanced picture of the causal effect of Airbnb presence on local housing demand.

Christian Tien, M Xiao, I Harkins, W Styles

Found in the (Random) Forest