Extremes of Shepp statistics for the Wiener process (Q2271712): Difference between revisions

Let \(W(t)\) be a standard Wiener process. Define \(X(t,s)=W(t+s)-W(t)\) and \(Y(t)=\max_{0\leq s\leq1}{X(t,s)}\). Let \(M_T=\max_{t\in[0,T]}Y(t)\) be the maximum up to time \(T\) of \(Y(t)\). The author deals with the asymptotic behaviour of \(\text{Pr}(M_T>u)\), the probability of high level excursions of \(Y(t)\) as \(u\to\infty\) and obtains the limiting distribution of \(M_T\) as \(T\to\infty\). The main results are the following theorems. Theorem 1. If \(Tu^2\to\infty\) and \(Tu^2\psi(u)\to0\) as \(u\to\infty\), then \(\text{Pr}(M_T>u)=HTu^2\psi(u)(1+o(1))\), where \(\psi(u)\) is the tail of the standard normal distribution function and \(H\) is a constant. Theorem 2. For any fixed \(x\) and \(T\to\infty\), the following relation holds: \[ \text{Pr}\left( \max_{(t,s)\in[0,T]\times[0,1]}a_T(W(t+s)-W(t)-b_T)\leq x\right)= e^{-e^{-x}}+o(1), \] where \(a_T=\sqrt{2\ln{T}}\), \(b_T=\sqrt{2\ln{T}}+(\ln{H}+(1/2)(\ln{\ln{T}}-\ln\pi))/\sqrt{2\ln{T}}\). The expression for the constant \(H\) is given. A similar result for standardized Wiener process increments is proposed by \textit{Z. Kabluchko} [``Extreme-value analysis of standardized Gaussian increments'', \url{http://arxiv.org/abs/0706.1849} (2008)]. There are also a number of papers on strong laws for increments of Wiener processes (see, for example, \textit{M. Csörgö} and \textit{P. Revesz} [Ann. Probab. 7, 731--737 (1979; Zbl 0412.60038)] and \textit{A. N. Frolov} [Theory Probab. Appl. 49, No. 3, 531--540 (2005); translation from Teor. Veroyatn. Primen. 49, No. 3, 601--609 (2004; Zbl 1098.60032)]).