Temporal Alignment and Covariate Matching for Robust Synthetic Controls

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.International Conference on Machine Learning, Artificial Intelligence and Data Science (ICMLAIDS 2026)www.srcmeetings.com17Conference Proceedings DOI: doi.org/10.47363/JAICC/ICMLAIDS2026/2026(5)17