A1274
Title: Synthetic control methods through predictive synthesis
Authors: Masahiro Kato - University of Tokyo (Japan) [presenting]
Akira Fukuda - Takushoku University (Japan)
Kosaku Takanashi - Keio University (Japan)
Kenichiro McAlinn - Temple University (United States)
Akari Ohda - University of Tokyo (Japan)
Masaaki Imaizumi - The University of Tokyo (Japan)
Abstract: Synthetic control (SC) methods have become a vital tool for causal inference in comparative case studies. The main idea of SC methods is to estimate the counterfactual outcomes of a treated unit using a weighted sum of observed outcomes of untreated units. Two novel methods are proposed for SC methods by synthesizing predictive densities. The first method synthesizes predictive densities using Bayesian predictive synthesis (BPS). The proposed method, the Bayesian Predictive SC (BPSC) method, has several advantages over frequentist SC methods. For instance, it can handle issues such as model misspecification and construct confidence intervals. Additionally, covariates can be used as predictors for outcomes by synthesizing them to predict counterfactual outcomes. The second method assumes a mixture model between the densities of treated and untreated units, and SC weights are estimated by density matching. The SC weights can be estimated by matching the higher moments of the treated unit and a weighted sum of untreated units. Using this method, the mean squared error of the counterfactual prediction in experiments is successfully minimized.
