Graph transformation can be used to implement stochastic simulation of dynamic
systems based on semi-Markov processes, extending the standard approach based
on Markov chains. The result is a discrete event system, where states are graphs,
and events are rule matches associated to general distributions, rather than just
exponential ones. We present an extension of this model, by introducing a
hierarchical notion of event location, allowing for stochastic dependence of higher-
level events on lower-level ones.