On the exact asymptotic behaviour of the distribution of the supremum in the ``critical'' case (Q5953877)

Let \(X_1,X_2,\dots\) be i.i.d. random variables with distribution \(F\) having an absolutely continuous component. Set \(S_0 = 0\), \(S_n =X_1+\cdots+X_n\), \(n\geq 1\). Let \(S_n\to -\infty\) a.s. so that \(M= \max(S_n, n\geq 0)<\infty\) a.s. In a previous paper [ibid. 52, No. 3, 301-311 (2001; Zbl 0990.60051)] the author studied the tail behaviour of the distribution of \(M\) under the condition that \(F(x,\infty)/G(x,\infty)\to\lambda < \infty\) as \(x\to \infty\) where \(G\) is a distribution belonging to \({\mathcal S}(\gamma)\) -- defined there -- with \(\gamma > 0\) such that \(\varphi(s)=\int \text{exp}(\gamma x)F(dx)\neq 1\), \(\text{Re} s =\gamma\), and also under a more complicated condition involving \(G\) when \(\varphi(\gamma)=1\) and \(\int x^2\text{exp}(\gamma x) F(dx) < \infty\). Here the tail behaviour is studied when \(\varphi(\gamma)=1\), \(\int x^2 \text{exp}(\gamma x) F(dx) = \infty\) and \(\int|x|\text{exp}(\gamma x) F(dx) < \infty\).