1: \begin{abstract}

2:   We consider inference on a regression coefficient under a constraint on the

3:   magnitude of the control coefficients. We show that a class of estimators

4:   based on an auxiliary regularized regression of the regressor of interest on

5:   control variables exactly solves a tradeoff between worst-case bias and

6:   variance. We derive ``bias-aware'' \acp{CI} based on these estimators, which

7:   take into account possible bias when forming the critical value. We show that

8:   these estimators and \acp{CI} are near-optimal in finite samples for \acl{MSE}

9:   and \ac{CI} length. Our finite-sample results are based on an idealized

10:   setting with normal regression errors with known homoskedastic variance, and

11:   we provide conditions for asymptotic validity with unknown and possibly

12:   heteroskedastic error distribution. Focusing on the case where the constraint

13:   on the magnitude of control coefficients is based on an $\ell_p$ norm

14:   ($p\ge 1$), we derive rates of convergence for optimal estimators and \acp{CI}

15:   under high-dimensional asymptotics that allow the number of regressors to

16:   increase more quickly than the number of observations.

17: \end{abstract}

18: