Benchmarking Counterfactual Prediction in Epidemic Time Series with Time-Varying Interventions

A new large-scale benchmark for counterfactual prediction in epidemic time series has been developed, addressing the need for realistic datasets with observable counterfactual outcomes. This benchmark supports both static and time-varying interventions in single and multi-policy settings, utilizing a calibrated agent-based model informed by real-world data across over 150 U.S. counties. It facilitates the evaluation of various causal inference methods, demonstrating significant performance disparities and underscoring the complexities involved in time-series causal reasoning, which is critical for practitioners working on causal inference in dynamic environments.