부산대학교 도서관

We introduce a variant of the Hawkes-fed birth-death process, in which the conditional intensity does not increase at arrivals, but at departures from the system. Since arrivals cause excitation after a delay equal to their lifetimes, we call this a delayed Hawkes process. We introduce a general family of models admitting a cluster representation containing the Hawkes, delayed Hawkes and ephemerally self-exciting processes as special cases. For this family of models, as well as their nonlinear extensions, we prove existence, uniqueness and stability. Our family of models satisfies the same FCLT as the classical Hawkes process; however, we describe a scaling limit for the delayed Hawkes process in which sojourn times are stretched out by a factor $\sqrt T$, after which time gets contracted by a factor $T$. This scaling limit highlights the effect of sojourn-time dependence. The cluster representation renders our family of models tractable, allowing for transform characterisation by a fixed-point equation and for an analysis of heavy-tailed asymptotics. In the Markovian case, for a multivariate network of delayed Hawkes birth-death processes, an explicit recursive procedure is presented to calculate the $d$th-order moments analytically. Finally, we compare the delayed Hawkes process to the regular Hawkes process in the stochastic ordering, which enables us to describe stationary distributions and heavy-traffic behaviour.
Comment: 38 pages, 1 figure