Bayesian Synthetic Control Methods with Spillover Effects: Estimating the Economic Cost of the 2011 Sudan Split

Shosei Sakaguchi,Hayato Tagawa

DOI: https://doi.org/10.48550/arXiv.2408.00291

2024-08-01

Abstract:The synthetic control method (SCM) is widely used for causal inference with panel data, particularly when there are few treated units. SCM assumes the stable unit treatment value assumption (SUTVA), which posits that potential outcomes are unaffected by the treatment status of other units. However, interventions often impact not only treated units but also untreated units, known as spillover effects. This study introduces a novel panel data method that extends SCM to allow for spillover effects and estimate both treatment and spillover effects. This method leverages a spatial autoregressive panel data model to account for spillover effects. We also propose Bayesian inference methods using Bayesian horseshoe priors for regularization. We apply the proposed method to two empirical studies: evaluating the effect of the California tobacco tax on consumption and estimating the economic impact of the 2011 division of Sudan on GDP per capita.

Econometrics

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve

This paper aims to address the issues of the Synthetic Control Method (SCM) when dealing with spillover effects. Specifically:

1. **Background and Challenges**:
- The Synthetic Control Method (SCM) is a widely used method for causal inference in panel data, especially when the number of intervention units is very small. SCM assumes the Stable Unit Treatment Value Assumption (SUTVA), which means that potential outcomes are not affected by the treatment status of other units.
- However, in reality, interventions not only affect the treated units but also the untreated units, which is known as spillover effects.

2. **Main Contributions**:
- This paper proposes a new panel data analysis method that extends the standard Synthetic Control Method to account for spillover effects, allowing for the simultaneous estimation of treatment effects and spillover effects.
- By introducing a spatial autoregressive panel data model to capture spatial correlations between different units, the paper relaxes the SUTVA assumption.
- A Bayesian inference method is proposed, utilizing Bayesian horseshoe priors for regularization to improve the accuracy of parameter estimation.

3. **Empirical Applications**:
- The method is applied to two empirical studies: evaluating the impact of California's tobacco tax on consumption and estimating the impact of Sudan's 2011 split on per capita GDP.
- The results show that California's tobacco tax not only reduced tobacco consumption in California but also decreased tobacco consumption in other states. Additionally, South Sudan's independence led to a decrease of about 20% in Sudan's per capita GDP, with a cumulative reduction of 72% from 2011 to 2020.

Through these improvements, the paper addresses the limitations of traditional Synthetic Control Methods in handling spillover effects, enhancing the accuracy of causal inference.