Siwon Ryu
Siwon Ryu

IconWorking Papers

“Causal Effects of Treatments with Network Changes” (Job market paper)

Abstract: Recent empirical studies emphasize the importance of indirect, or spillover, effects in program evaluation. Most studies assume that the underlying network is exogenous, fixed, or unaffected by the intervention. However, empirical evidence indicates that the treatment can also have significant network effects. This paper studies the identification and estimation of causal treatment effects while explicitly considering possible causal changes in the network resulting from a program. The main finding is the decomposition of the causal effects into two distinct components: the treatment effect when the network remains unchanged and the effect when the treatment alters only the network structure (network effect). This result enhances our understanding of policy/program mechanisms by considering counterfactual scenarios where the network is either altered or remains unchanged due to the treatment. The proposed method applies to both randomized experiments and quasi-experimental designs with parallel trends. A estimation procedure for causal effects and their decomposition is proposed, and its performance is evaluated through Monte Carlo simulations. The methodology is illustrated using data from a program offering savings accounts. The empirical results show that total direct effect is small due to offsetting positive treatment effects and negative network effects.

“Local Average Treatment Effects with Imperfect Compliance and Interference” (Submitted, Journal of Econometrics)

Abstract: This study analyzes the identification and estimation of causal effects in situations where units interact and treatment is endogenous due to imperfect compliance. In cases where units do not interact, monotonicity in potential treatments identifies local average treatment effects (LATE). When units interact, monotonicity can still apply, but additional restrictions on potential treatments, such as one-sided noncompliance or personalized encouragement, are typically required. This paper generalizes these restrictions into a weaker concept of monotonicity and provides a unified framework for this context. Direct and indirect LATEs are identified under strictly weaker restrictions on potential treatments compared to existing approaches, but with the assumption of an additional exclusion restriction for the endogenous treatment. A parametric estimator for causal effects is introduced, and its performance is evaluated through simulations. The estimator also assesses biases in existing methods when various underlying assumptions are violated. The estimation procedure is illustrated using an experimental study in Kenya, which provided access to a savings account.

“Causal Mechanism with Interference”

Abstract: This study examines identification and estimation in the presence of social interactions. Potential outcome depends on both an individual’s own treatment and the exposures generated by others’ treatment statuses and the underlying network structure. When the network itself is impacted by treatment, the individual’s treatment can also alter these exposures. In such cases, the treatment effect can be decomposed into direct and indirect components. The direct effect represents the causal impact on the outcome when exposures are fixed, while the indirect effect captures the influence of changes in the exposure distribution due to the individual’s own treatment. As a result, observed exposures can be viewed as mediators of the individual’s treatment effect. I employ a causal mediation framework to identify and decompose these effects and propose an estimation method. The performance of the estimator is then evaluated through Monte Carlo simulations, followed by an empirical application examining the impact of coeducational high schools on academic performance.

IconWork in Progress

“Heterogeneous Parental Labor Supply Responses to Children’s Severe Health Shocks: Evidence from Administrative Data from South Korea” (with Jungmin Lee and Hyuncheol Bryant Kim)

Abstract: This study examines the impact of a child’s cancer diagnosis on parents’ labor supply. Using data from South Korea’s National Health Insurance System, we find heterogeneous effects based on gender, parents’ relative income, and employment type. Specifically, we observe that mothers experience a significant decline in employment and income following their child’s cancer diagnosis, while fathers’ employment and income remain relatively unchanged. The decrease in fathers’ labor supply is only observed among those with lower income compared to their spouse, whereas mothers’ employment decreases regardless of relative income. Furthermore, we find that self-employed mothers are less affected compared to wage workers, while the effects on fathers are generally minimal.

“Financial Development and Trade Network Formation” (with Daeeun Bae)