1: \begin{abstract} 

2:   \begin{spacing}{1}

3:   We propose a method to conduct uniform inference for the optimal value

4:   function, that is, the function that results from optimizing an objective

5:   function marginally over one of its arguments. Marginal optimization is not

6:   compactly differentiable as a map between the spaces of objective and value

7:   functions, which is problematic because standard inference methods for

8:   nonlinear maps usually rely on compact differentiability.  However, we show

9:   that the map from objective function to uniform test statistics applied to the

10:   value function~--- specifically, Kolmogorov-Smirnov or Cram\'er-von Mises

11:   statistics~--- are directionally differentiable. We establish consistency and

12:   weak convergence of nonparametric plug-in estimates of the test statistics.

13:   For practical inference, we develop detailed resampling techniques that

14:   combine a bootstrap procedure with estimates of the directional derivatives.

15:   In addition, we establish local size control of tests which use the resampling

16:   procedure.  Monte Carlo simulations assess the finite-sample properties of the

17:   proposed methods and show accurate empirical size of the procedures. Finally,

18:   we apply our methods to the evaluation of a job training program using bounds

19:   for the distribution function of treatment effects.  

20:   \end{spacing}

21: \end{abstract}

22: