Working Paper – The Bayesian Synthetic Control

Elias Tuomaala. (2019) “The Bayesian Synthetic Control: Improved Counterfactual Estimation in the Social Sciences through Probabilistic Modeling.” Arxiv Open Access. [link]

In this working paper I propose a novel statistical approach, the Bayesian Synthetic Control (BSC), to address a frequently encountered problem in econometrics. BSC can be used to estimate the impact of one-off policy reforms (ie. tax hikes, trade liberalization, Brexit) on individual targeted countries or cities. The method is inspired by earlier work, namely the Synthetic Control Method (SCM) and its frequentist variants (GSC, ASCM). The previous methods were unable to generate confidence intervals for their estimation results – BSC lacks this and certain other limitations.

I developed the Bayesian Synthetic Control as my undergraduate senior thesis at Harvard College, under the invaluable advising of Dr. Rahul Dave from the Harvard School of Engineering and Applied Sciences. The project also greatly benefited from the guidance of Prof. Matthew Blackwell from the Harvard Department of Government.

The bayessynth Python Library

Elias Tuomaala. (2020) “bayessynth: BSC Models in Python.” Python library. Github repository.

In this (currently unmaintained) library, I provide a Python implementation of the Bayesian Synthetic Control (BSC). It can be used, more or less out-of-the-box, to fit the BSC model on any appropriate dataset. It includes tools for running Markov Chain Monte Carlo (MCMC) sampling to estimate the model as well as for analyzing and visualizing the sampling outputs.