Haomin Yu

We develop a Gaussian approximation to the finite-sample distribution of the classical ridge regression estimator. The approximation accommodates heteroskedasticity and autocorrelation and motivates new average- and worst-case-risk methods for selecting the regularization parameter.

We study how to summarize heterogeneous policy effects with a small number of archetypes. A decision-theoretic formulation leads to weighted k-means clustering of policy-effect estimates and yields Bayes and approximately minimax procedures.