A0536
Title: A L-infinity norm counterfactual and synthetic control approach
Authors: Le Wang - Virginia Tech (United States) [presenting]
Abstract: The synthetic control (SC) method is a widely used tool for policy evaluation and causal inference. It often employs sparse regularization techniques to address challenges arising from the number of pre-treatment periods being modest relative to the number of control units. While sparsity enhances transparency, it can increase sensitivity to estimator variability, amplify bias, and reduce reliability when control units are drawn from a common distribution. To address these limitations, a novel SC extension is proposed incorporating infinity norm regularization. The approach introduces a denser weighting scheme that assigns weights to a broader set of control units while constraining the largest weight. This reduces reliance on unrepresentative units and enhances robustness while maintaining the same level of desirable match between treated and synthetic units. An interior point method is developed to efficiently solve the infinity-norm constrained problem. A statistical theoretical framework is built to derive the asymptotic distribution of the proposed estimators under conditions of weak dependence. Monte Carlo simulations demonstrate robust finite-sample performance, while applications to two real studies show the utility of the methods in a highly volatile stock market where conventional SC methods relying on a few control units are more susceptible to shocks and biased estimates.
