跳到主要內容區

 

:::

【統計學系演講】Testing for rotational symmetry under elliptical directions using multivariate signs

日期 : 2026-11-18 單位 : 統計系

主講人:Dr. Gaspard Bernard (柏嘉偲 助研究員) 中央研究院統計科學研究所

題      目:Testing for rotational symmetry under elliptical directions using multivariate signs

時      間:民國115年11月23日 (星期一) 下午1:30 

地      點:國立政治大學逸仙樓050101教室

摘      要:

Given a collection of random vectors, we consider the problem of testing for rotational symmetry of the underlying distribution. It is well known that, under elliptical symmetry, testing for rotational symmetry is equivalent to testing whether a scatter parameter is a multiple of the identity matrix. We consider the more general model of random vectors with elliptical directions introduced by Randles and identify several scenarios in which testing for rotational symmetry is still equivalent to testing whether the scatter parameter is a multiple of the identity. Among these, we place particular emphasis on scenarios in which a rather general form of dependence between observations may be present, without requiring mixing conditions. We study, under these new assumptions, the behavior of the classical spatial-sign test and show that, under certain mild conditions, the test is asymptotically valid and has the same local asymptotic power as in the classical elliptical setting. We then show that, perhaps surprisingly, the spatial-sign test enjoys local asymptotic optimality properties for testing sphericity when the underlying distribution is strongly heavy-tailed. We also illustrate our procedure on a signal-detection problem involving a heavy-tailed, time-dependent financial returns dataset that appears compatible with our model assumptions.

瀏覽數: