Self-normalized moderate deviations and lils (Q1805758)

Let \(\{X_n\}\) be a sequence of i.i.d. \(\mathbb R^d\)-valued zero mean random variables either in the domain of attraction of a Gaussian law or in the domain of attraction of a stable law. A partial large deviation principle (PLDP) of speed \(a_n = o(n) \) is established for random vectors \(( T_n, U_{n,p})\) defined for \(p > 1\) by \[ T_n = a_n^{-1} z_n^{-1} \sum_{i=1}^n X_i, \quad U_{n,p} = a_n^{-1} z_n^{-1} \sum_{i=1}^n \|X_i \|^p \] where \( z_n \to \infty \) and \(\|\cdot \|\) denotes a norm on \(\mathbb R^d\). The proofs are based on an asymptotic expansion of moment generating function. As an application of the PLDP, law of iterated logarithm type-results are developed, which are of independent interest.