Complete corrected diffusion approximations for the maximum of a random walk (Q997958)

Let \(\{ X_{n},n\geq 1\} \) be a sequence of independent identically distributed random variables and let \(S_{0}=0,\;S_{n}=\sum_{j=1}^{n}X_{j}\), \(n\geq 1\); \(M=\max \{ S_{n},n\geq 0\}\). For \(x>0\), \(\tau (x)=\inf \{ n\geq 1\), \(S_{n}\geq x\}\); \(\{ M\geq x\} =\{ \tau (x)<\infty \} .\) In this paper, \( EX_{i}=-\mu\), \(\mu >0\) and \(E\exp \{ \theta X_{i}\} <\infty \) for \(\theta \) in the neighborhood of origin. For \(\theta \) of this kind define \( \psi ( \theta ) =\log E( \exp ( \theta X_{1}) ) \) and the probability measure \(P_{\theta }\)\ such that \[ P_{\theta }( A) =E( \exp ( \theta S_{n}-n\psi (\theta ) ) \mathbf{1}_{A}) \] for \(A\in \delta ( S_{j},\;0\leq j\leq n)\). The aim of the paper is to give corrected approximations to \(P_{\theta }( M>x) \) as \(\theta \uparrow \infty .\) Due to the identity of Wald's likelihood ratio \[ \begin{aligned} P_{\theta }( \tau (x)>x) &= E_{\theta }\exp \{ -( \theta _{1}-\theta _{0}) S_{\tau (x)}\} \\ &= \exp \{ -( \theta _{1}-\theta _{0}) x\} \cdot E_{\theta _{2}}\exp \{ -( \theta _{1}-\theta _{0}) R( x) \} \end{aligned}, \] where \(R( x) =S_{\tau (x)}-x\) is the so-called ``overshoot'' at the level \(x\) for \(\theta_{0}\) in some interval of the form \(( -\eta ,0)\) and there exists a positive \(\theta _{1}\) such that \(\psi ( \theta _{0}) =\psi ( \theta _{1})\). Set \(\Delta =\theta _{1}-\theta _{0}\). Since \(X_{1}\) is strongly nonlattice, the renewal theory can be applied to \(\Delta <\eta\). Hence, \[ P_{\theta _{0}}( M>x) =\exp ( -\Delta x) \cdot E_{\theta _{1}}\exp \{ -\Delta R( \infty ) \} +o( \exp ( -( \Delta +r) x) ) \] for some \(r>0\). Theorem 1 of the paper states that, if \(X_{1}\) has exponential moments and is strongly nonlattice, then \[ r( \Delta ) =\log E_{\theta _{1}}\exp \{ -\Delta R( \infty ) \} \] admits an analytic extension in the neighborhood of the origin on the complex plane. For \(X_{1}\) symmetric with a continuous distribution function full asymptotic expansions are obtained for the expected ladder height \(E_{\theta _{1}}S_{\tau _{+}}\) where \(\tau _{+}=\inf \{ n\geq 1,\;S_{n}>0\}\). That is Theorem 2. For nonnegative \(\theta \in\mathbb{N}\) and \(b\in\mathbb{R}\) denote \[ \rho ( \theta ,b) =\log E_{\theta }\exp ( -bR( \infty ) ). \] Theorem 3 states that, if \(X_{1}\) has exponential moments and is strongly nonlattice, then, for \(0<\theta \in\mathbb{N}\) and \(b>0\), \[ \rho ( \theta ,b) =\frac{1}{2\pi }\int \limits_{-\infty }^{\infty }\frac{-b}{( b+i\lambda ) i\lambda }\log \left( \frac{2( \gamma ( \theta ) -\gamma ( \theta +i\lambda ) ) }{\lambda ( \lambda -2i\varphi ^{\prime }( \theta ) ) } \right) d\lambda \] where \(\varphi (\theta )=E\exp ( \theta X_{1})\). The asymptotics as \(b\downarrow 0\) of such integrals developed in the next sections of the paper is applied in proving Theorem 4, namely, under the same conditions on \(X_{1,}\) for all \(n\geq 1\), the derivative \[ K_{n}( \theta ) \equiv ( -1) ^{n}\left. \frac{\partial ^{n}}{\partial b^{n}}S( \theta ,b) \right| _{b=0} \] initially defined in \(( 0,v) \) for \(v>0\) can be extended to be an analytic function on a disc containing the origin in the complex plane. An implication of Theorem 4 is that it can be directly applied to complete asymptotics of the steady-state mean of the waiting time sequence, namely for \(E_{\theta _{0}}M( \equiv E_{\theta }W_{\infty }) ,\) as well as to obtain the full asymptotic expansion for \(K_{n}( \cdot ) \) in the neighborhood of zero (see, Section b.\(2\)). In section \(3\), the short-time asymptotics for the Cauchy process is considered, and in Section \(4\), their possibilities of reducing the corresponding integrals to that used in the general case are discussed. Section \(5\) provides a detailed description of the arguments used to compute the coefficients of the expansion. Section \(6\) presents how the statements of the three previous sections can be applied in obtaining an asymptotic expansion in powers of \(\Delta \) for \(r( \Delta ) \) and the asymptotic expansion in powers of \(\theta\) for the cumulants of \(R( \infty ) \) by the measure \(P_{\theta }.\) Section \(7\) contains technical proofs of Theorems 1--4, formulated in section \(2\). In the introduction various fields of problems and random processes related to approximations of \(M\) are discussed. The result of the previous authors are mentioned and compared with that of the paper. The list of references contains \(17\) positions.