Last exit times for random walks (Q1120191)

Let \(\{S_ n\), \(n\geq 0\}\) be a random walk which drifts to \(+\infty\), i.e. \(S_ 0=0\), \(S_ n=\sum^{n}_{1}X_ j\), \(n\geq 1\), where the X's are i.i.d. and \[ \sum^{\infty}_{1}n^{-1}P(S_ n\leq 0)<\infty,\quad \sum^{\infty}_{1}n^{-1}P(S_ n>0)=\infty. \] Parallel to first exit times, the author defines last exit times \(L(x)=\sup \{n\geq 0:\) \(S_ n\leq x\}\), and the last exit ladder point set \[ \cup_{x}\{(L(x),S_{L(x)})\}=\{(n,S_ n):\quad \forall m>n,\quad S_ m>S_ n\}=\{(E_ k,Z_ k),\quad k\geq 1\}, \] where \(0<E_ 1<E_ 2<...<E_ k<..\). Then it is shown that \(\{(E_ k,Z_ k)\), \(k\geq 1\}\) is a modified 2-dimensional renewal process on [0,\(\infty)\times (- \infty,\infty)\), i.e. \((E_ 1,Z_ 1)\), \((E_{k+1}-E_ k\), \(Z_{k+1}- Z_ k)\), \(k\geq 1\), are independent, and \((E_{k+1}-E_ k\), \(Z_{k+1}- Z_ k)\), \(k\geq 1\), are identically distributed, and the marginal process \(\{E_ k\), \(k\geq 1\}\) is a stationary renewal process. For some special cases, the asymptotic behavior of \(P(L(x)=n)\) and \(P(E_ k=n)\) as \(n\to \infty\) is discussed.