Logarithmic asymptotics for the supremum of a stochastic process (Q1413673): Difference between revisions

Let \(\{W_t\}\) be a stochastic process, and let \(Q= \sup_{t\geq 0} W_t\). The paper investigates the tail asymptotics of \({\mathbf P}\{Q> q\}\) as \(q\) becomes large. Assume that the process \(\{W_t\}\) satisfies a restricted form of the large deviation principle: for some scaling functions \(a\), \(\nu\) regularly varying with indices \(A> 0\), \(V> 0\), respectively, the limit \[ \lim_{t\to\infty}\nu(t)^{-1}\log{\mathbf P}\{a(t)^{-1} W_t> c\}= -J(c) \] exists for all \(c\geq 0\). Under additional technical assumptions on \(\{W_t\}\), it is proved that \[ \lim_{q\to\infty} h(q)^{-1}\log{\mathbf P}\{Q> q\}= -\inf_{c> 0} c^V J(c^{-A}) \] with \(h=\nu(a^{-1})\).