1: \begin{abstract}

2: We present a novel inference approach which we call Sample Out-of-Sample (or

3: SOS) inference. Our motivation is to propose a method which is well suited for

4: data-driven stress testing, in which emphasis is placed on measuring the

5: impact of (plausible) out-of-sample scenarios on a given performance measure

6: of interest (such as a financial loss). The methodology is inspired by

7: Empirical Likelihood (EL), but we optimize the empirical Wasserstein distance

8: (instead of the empirical likelihood) induced by observations. From a

9: methodological standpoint, our analysis of the asymptotic behavior of the

10: induced Wasserstein-distance profile function shows dramatic qualitative

11: differences relative to EL. For instance, in contrast to EL, which typically

12: yields chi-squared weak convergence limits, our asymptotic distributions are

13: often not chi-squared. Also, the rates of convergence that we obtain have some

14: dependence on the dimension in a non-trivial way but which remains controlled

15: as the dimension increases.

16: 

17: \end{abstract}