Interested parties are welcome to attend.
Presentation Title: SYNTHETIC DECOMPOSITION FOR COUNTERFACTUAL PREDICTIONS
(with Nathan Canen)
Abstract:
Producing predictions for a new policy is one of the most important, yet challenging, problems in empirical research. A common method in the literature is to decompose the data generating process from a source population into a policy-invariant structure and source-specific covariates and transfer the invariant structure to the target population. In this paper, we generalize this approach to a setting where there are multiple source populations (e.g., multiple regions/countries/markets that have been subject to similar policies of interest in the past). We propose a novel method of constructing a synthetic policy-invariant structure from these source populations to generate predictions for a new policy in the target population. For this, we formulate a policy-invariance condition, and develop data-dependent weights for the synthetic structure so that it is as close as possible to satisfying the invariance condition. We develop a general procedure to construct asymptotic confidence intervals for counterfactual predictions and prove its asymptotic validity.
Contact Marinho Bertanha for information.