Dynamic Synthetic Control with Reinforcement Learning for Policy Optimization

Cerqua et al. (2023)’s ML Control Method relies on static pre-treatment periods, ignoring policy evolution. This idea treats policy optimization as a reinforcement learning problem: states sequentially adjust policies (e.g., tax rates) while a synthetic control agent learns optimal interventions from counterfactual outcomes. Unlike Chu et al.’s (2023) continual causal inference—which focuses on data drift—we target policy drift (e.g., mid-term legislative amendments). By integrating Wang et al.’s (2024) heterogeneity-robust estimators as reward signals, we address Pernia’s (2023) critique that political science lacks dynamic causal tools. This could revolutionize adaptive governance (e.g., real-time pandemic response).

References:

1. Causal inference and policy evaluation without a control group. Augusto Cerqua, M. Letta, Fiammetta Menchetti (2023).
2. Advances in Difference-in-differences Methods for Policy Evaluation Research. Guangyi Wang, Rita Hamad, Justin S. White (2024). Epidemiology.
3. Continual Causal Inference with Incremental Observational Data. Zhixuan Chu, Ruopeng Li, S. Rathbun, Sheng Li (2023). IEEE International Conference on Data Engineering.
4. Causal inference in political science research: global trends and implications on Philippine political scholarship. Ronald A. Pernia (2023). Asian Journal of Political Science.