Some extensions of the LIL via self-normalizations (Q756248): Difference between revisions

The authors study general versions of the Law of the iterated logarithm for partial sums of independent random variables \(X_ i\) with E \(X_ i=0\) and possibly infinite variance. Self-normalization is used, i.e. the variance term \(\sigma^ 2_ n\) in the classical LIL normalizer \(\sqrt{\sigma^ 2_ n \log \log \sigma^ 2_ n}\) is replaced by the estimator \(V^ 2_ n=X^ 2_ 1+...+X^ 2_ n.\) Three main results are presented, the first being an upper class result in the case of symmetric r.v.'s, extending a bounded LIL of Marcinkiewicz. Secondly, an extension of the Kolmogorov-Erdős test to i.i.d. symmetric random variables with finite variance is given, identifying the class of functions \(\phi\) for which \(S_ n\leq V_ n\phi (n)\) eventually. Finally, the bounded LIL of Marcinkiewicz is generalized to the non-symmetric case with an unusual normalization, namely a mixture of classical and self-normalization. As a byproduct a fairly short and elementary proof of the classical LIL is obtained.