Hyperedge-Triggered Hawkes processes for diagnosing higher-order temporal dependence in event streams
We study whether the recent activity of an entire group adds information about subsequent events on its members. We introduce a hyperedge-triggered Hawkes process in which a group completes when every member has an event within a fixed time window. Its most recent completion defines a bounded, exponentially decaying contribution to the intensities of the same members. The completion feature contains a joint history effect that cannot be represented by an additive collection of proper-subset history terms. We prove finite-horizon nonexplosion and give a sufficient condition for a finite-mean stationary version. The most recent completion also yields an exact compensator and a concave coefficient log-likelihood for fixed candidates and time scales. Estimation combines source-allocation EM with coordinate maximization. A study of 1100 simulated paths examines overlapping three-node groups and time-varying backgrounds. The three-node component provides positive mean held-out log-likelihood increments in the positive-signal scenarios after adjustment for two-node completion effects. Applications to three neural recordings assess completion effects across fixed time windows and background specifications. The resulting comparison quantifies the predictive contribution of a specified group-completion mechanism.
Authors
- Zihan Xu (ORCID: https://orcid.org/0009-0003-6631-2971)
Institutions
- Qingdao University (CN)
Publication Details
- Journal
- Journal of Statistical Computation and Simulation
- Published
- 2026-11-13
- DOI
- https://doi.org/10.1080/00949655.2026.2762892
- Primary Topic
- Point processes and geometric inequalities
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Harvard University