Self-normalized processes: exponential inequalities, moment bounds and iterated logarithm laws. (Q1879810)

Exponential and \(L_p\) bounds for a martingale divided by the square root of its quadratic variation are established and used in the derivation of laws of the iterated logarithm for these objects. The article contains a wealth of results that subsume previous work on the subject by several authors. Here is one of the theorems as an example: Let \(\{ X_n \}\) be a martingale difference sequence and let \(V_n^2=\sum_{i=1}^n X_i^2\). Assume \(|X_n|\leq m_n\) a.s., where \(m_n\) is \({\mathcal F}_{n-1}\)-measurable (\(n\in{\mathbb N}\)), that \(V_n\to\infty\) a.s. and that \(m_n/[V_n(\log\log (V_n \vee e^2))^{-1/2}]\to 0\) a.s. Then, the cluster set of the sequence \(\left\{ \sum_{i=1}^n X_i / [V_n(\log\log (V_n \vee e^2))^{1/2}] \right\}\) is a.s. the interval \([-\sqrt{2},\sqrt{2}]\). This particular theorem is related to results of \textit{P. S. Griffin} and \textit{J. D. Kuelbs} [Ann. Probab. 17, 1571-1601 (1989; Zbl 0687.60033)], and, to some extent, also to \textit{E. Giné} and \textit{D. M. Mason} [J. Theor. Probab. 11, 351--370 (1998; Zbl 0919.60047)].