Hawkesprocess

A Hawkes process is a type of point process where the rate of events is not constant but depends on the past history of the process. This means that an event occurring can influence the probability of future events. Specifically, in a Hawkes process, an event tends to trigger more events in its immediate temporal neighborhood. This self-exciting property is a key characteristic. The intensity of the process at any given time is a sum of a background intensity and contributions from previous events, each weighted by a kernel function that decays over time.

The mathematical formulation typically involves a time-varying intensity function, often denoted by lambda(t). This lambda(t) is composed of a baseline rate plus a sum of kernel functions, K(t - t_i), for each past event time t_i. The kernel function K determines how much influence a past event has on the current intensity. Common choices for K include exponential decay or power-law decay. Hawkes processes are widely used to model phenomena exhibiting clustering, where events are not uniformly distributed in time but tend to occur in bursts.

Applications of Hawkes processes are found in diverse fields. They are used in seismology to model earthquake aftershocks, where a large earthquake increases the probability of smaller earthquakes in the vicinity. In finance, they can model the clustering of trading activity or the propagation of price changes. Network security uses Hawkes processes to detect and analyze patterns of malicious activity, as attacks can often trigger subsequent attacks. Other areas include neuroscience, where they model neuronal firing, and social sciences, to understand the spread of information or the occurrence of riots. The ability to capture temporal dependencies makes them powerful tools for analyzing and predicting event sequences.