We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

math.ST

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Mathematics > Statistics Theory

Title: Sample Out-Of-Sample Inference Based on Wasserstein Distance

Abstract: We present a novel inference approach that we call Sample Out-of-Sample (or SOS) inference. The approach can be used widely, ranging from semi-supervised learning to stress testing, and it is fundamental in the application of data-driven Distributionally Robust Optimization (DRO). Our method enables measuring the impact of plausible out-of-sample scenarios in a given performance measure of interest, such as a financial loss. The methodology is inspired by Empirical Likelihood (EL), but we optimize the empirical Wasserstein distance (instead of the empirical likelihood) induced by observations. From a methodological standpoint, our analysis of the asymptotic behavior of the induced Wasserstein-distance profile function shows dramatic qualitative differences relative to EL. For instance, in contrast to EL, which typically yields chi-squared weak convergence limits, our asymptotic distributions are often not chi-squared. Also, the rates of convergence that we obtain have some dependence on the dimension in a non-trivial way but remain controlled as the dimension increases.
Subjects: Statistics Theory (math.ST)
Cite as: arXiv:1605.01340 [math.ST]
  (or arXiv:1605.01340v4 [math.ST] for this version)

Submission history

From: Yang Kang [view email]
[v1] Wed, 4 May 2016 16:37:05 GMT (36kb)
[v2] Tue, 17 May 2016 02:36:03 GMT (37kb)
[v3] Wed, 24 Aug 2016 23:47:05 GMT (38kb)
[v4] Tue, 18 Feb 2020 01:04:55 GMT (1233kb,D)

Link back to: arXiv, form interface, contact.