Double and Single Descent in Causal Inference with an Application to High-Dimensional Synthetic Control

Motivated by a recent literature on the double-descent phenomenon in machine learning, we consider highly over-parameterized models in causal inference, including synthetic control with many control units.