A0994
Title: Bayesian donor set selection in synthetic control method
Authors: Joungyoun Kim - University of Seoul (Korea, South) [presenting]
Johan Lim - Seoul National University (Korea, South)
Seul Lee - Seoul National Univerisity (Korea, South)
Abstract: The synthetic control method (SCM) is a methodology designed to assess the effect of a specific event by (i) finding a donor set that is not exposed to the event and (ii) predicting counterfactual outcomes of the event from the actual outcomes of a donor set. No doubt, the selection of an appropriate donor set is an important step in SCM, but it has rarely been researched in the literature. A new Bayesian hierarchical model is proposed, which incorporates a donor set selection process to the existing Bayesian synthetic control models. The new model relaxes the Dirichlet prior for the constrained regression coefficients in the existing Bayesian models as a hierarchical prior with Gamma variables and indicator variables for the donor set. A Markov chain Monte Carlo steps are developed to estimate the model. It is numerically shown how the inclusion of the donor set selection process improves the prediction of the counterfactual outcomes and, thus, the estimation of the effect of the event, compared to the existing Bayesian models without the selection of a donor set. The model is finally applied to the GDP trajectory of West Germany.
