Theory and applications of multivariate self-normalized processes (Q1045798)

The authors consider multivariate self-normalized processes in a general context. The studied processes are of the form \(A_n' C_n^{-1} A_n\) with a matrix normalization \(C_n^{-1},\) where \(C_n\) is a positive definite random matrix. An important tool to analyze \(C_n\) is its singular value decomposition and the corresponding symmetric square root. A parameterized family of exponential supermartingales is introduced to provide the general framework for the multivariate self-normalized processes. Then first, via the method of mixtures (or pseudo-maximization), some key exponential and moment inequalities are derived. In addition, boundary crossing probabilities for multivariate self-normalized processes and a Strassen-type law of the iterated are presented. The work extends earlier results of \textit{V. H. de la Peña, M. J. Klass} and \textit{T. L. Lai} [``Moment bounds for self-normalized martingales'', in: Giné, Evarist (ed.) et al., High dimensional probability II. 2nd international conference, Univ. of Washington, DC, USA, August 1--6, 1999. Boston, MA: Birkhäuser. Prog. Probab. 47, 3--11 (2000; Zbl 0967.60052); Ann. Probab. 32, No.~3A, 1902--1933 (2004; Zbl 1075.60014)] to the multivariate case.