Temporal Alignment and Covariate Matching for Robust Synthetic Controls

Abstract

 Synthetic control estimation under varying data regimes requires distinct strategies for counterfactual alignment. When direct controls are available, KD-tree–based nearest-neighbor search combined with entropy 
balancing enforces pre-treatment covariate similarity, enabling difference-in-differences weighting for robust causal inference. In contrast, absence of controls necessitates dynamic trajectory matching: Dynamic Time 
Warping identifies series with aligned temporal patterns, while Bayesian Structural Time Series synthesizes these into latent counterfactual paths via state-space modeling. This framework addresses imbalance, temporal 
misalignment, and structural uncertainty inherent in observational causal analysis.