Subexponential asymptotics of hybrid fluid and ruin models (Q1774229)

The authors consider the tail behaviour of the random variable \(V_{X+Y}^c := \sup_{t \in \mathbb{Z}} (X_t + Y_t - ct)\), where \(c\) is a constant and \(X = (X_t)_{t \geq 0}\) and \(Y = (Y_t)_{t \geq 0}\) are two independent stochastic processes. Specifically, \(Y\) is taken to be an integrated on-off process with peak rate \(r\), and the tail behaviour of \(V_{X+Y}^c\) is investigated for the three cases \(r > c\), \(r=c\) and \(r < c\), under some additional assumptions. In the first case \(r > c\), the generic random variable \(T_{\text{on}}\) of the on-periods of the source is of the form \(P(T_{\text{on}} > u) = \exp ( -L(u) u^\beta ) \), \(0 < \beta < 1\), where \(L\) is slowly varying at \(\infty\) and satisfies some further smoothness conditions, and \(X\) is standard Brownian motion. In both the second and the third case \(r = c\) and \(r < c\), respectively, \(T_{\text{on}}\) is supposed to be regularly varying and \(X\) is a centered Gaussian process with stationary increments, continuous sample paths, and a variance function \(t \mapsto \text{Var} (X_t)\) which is increasing, continuous, and regularly varying at 0 and at \(\infty\) with indices in \((0,2]\) and in \((0,2)\), respectively. Examples for \(X\) include fractional Brownian motion and integrated Gaussian processes. The paper concentrates on cases where the ``reduced-load equivalence'' \[ \lim_{u \to \infty} \frac{ P \left( \sup_{t \geq 0} (X_t + Y_t - ct) > u \right)}{ P \left( \sup_{t \geq 0} (Y_t - ct) > u \right)}=1 \] does not hold. For the cases considered, it is shown that the Gaussian process \(X\) contributes to the tail asymptotics of \(V_{X+Y}^c\) by its moderate deviations (case 1), large deviations (case 3), or oscillatory behaviour (case 2). The proofs include results on the tail behaviour of standard Brownian motion with drift and that of its running maxima process when sampled at certain subexponential times, independent of the Brownian motion.