Bayesian Matrix Completion Approach to Causal Inference with Panel Data

This study proposes a new Bayesian approach to infer binary treatment effects. The treats counterfactual untreated outcomes as missing observations and infers them by completing matrix composed of realized potential using data augmentation technique. We also develop tailored prior that helps in the identification parameters induces be approximately low rank. Posterior draws are simulated Markov Chain Monte Carlo sampler. While proposed is similar synthetic control methods other related methods, it has several notable advantages. First, unlike does not require stringent assumptions. Second, contrast non-Bayesian approaches, method can quantify uncertainty about inferences straightforward consistent manner. By means series simulation studies, we show our proposal better finite sample performance than existing approaches.