Efficiently Learning Synthetic Control Models for High-dimensional Disaggregated Data