Kirill Borusyak, University College London. "Non-Random Exposure to Natural Experiments: Theory and Applications"
Department webinar. Kirill Borusyak is an Assistant Professor at the University College London (UCL). He will present the paper: "Non-Random Exposure to Natural Experiments: Theory and Applications". Coauthored by Peter Hull.
We develop new tools for causal inference in settings where a natural experiment affects the treatment status of multiple observations jointly, to different extents. In such settings, researchers may construct instruments which combine exogenous shocks with pre-determined measures of shock exposure. Examples include shift-share instruments, instruments for spillovers in social and transportation networks, simulated eligibility instruments, and model-implied instruments. We show that the validity of these instruments requires a simple but non-standard recentering, derived from the specification of counterfactual shocks that might as well have been realized. We further show how this specification can be used for valid finite-sample inference and specification tests, and characterize the recentered instruments that are asymptotically efficient. Three applications illustrate our framework: we study the employment effects of high-speed rail with a market access instrument, the insurance coverage effects of Medicaid eligibility with a simulated instrument, and the local labor market effects of Chinese import penetration with a shift-share instrument.