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

Download:

Current browse context:

math

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: Ensemble minimaxity of James-Stein estimators

Abstract: This article discusses estimation of a multivariate normal mean based on heteroscedastic observations. Under heteroscedasticity, estimators shrinking more on the coordinates with larger variances, seem desirable. Although they are not necessarily minimax in the ordinary sense, we show that such James-Stein type estimators can be ensemble minimax, minimax with respect to the ensemble risk, related to empirical Bayes perspective of Efron and Morris.
Subjects: Statistics Theory (math.ST); Methodology (stat.ME)
Cite as: arXiv:2206.10856 [math.ST]
  (or arXiv:2206.10856v1 [math.ST] for this version)

Submission history

From: Yuzo Maruyama [view email]
[v1] Wed, 22 Jun 2022 05:56:12 GMT (39kb)

Link back to: arXiv, form interface, contact.