The asymptotic distribution of magnitude-Winsorized sums via self- normalization (Q910804)

The purpose of the paper is to present a new approach to tightness and weak convergence theorems for laws formed from constantly normalized partial sums. The approach consists of first deriving self-normalized (``studentized'') results and then comparing the self-normalizers with the constant normalizers. The final result is the knowledge of tightness or weak convergence of three sequences of laws, the self-normalized sums, the constantly normalized sums, and the ratio of the self-normalizers to the constant normalizers. Specifically the study is focused on the behaviour of the magnitude-Winsorized sums form from i.i.d. random variables that are symmetric about 0. The basic results include the following: 1. A characterization of when there exist norming constants so that the magnitude-Winsorized sums are asymptotically normal. 2. Identification of the precise class of subsequential limit laws that can and do arise. 3. Necessary and sufficient conditions for the magnitude-Winsorized sums to be attracted or partially attracted to a given (subsequential) limit law. 4. Necessary and sufficient conditions for stochastic compactness of the magnitude-Winsorized sums when normalized by constants. 5. A self-normalized or ``studentized'' limit theorem for the magnitude-Winsorized sums. An application of the methodology to trimmed sums is discussed to demonstrate its more general applicability as well as to illustrate its use.