Events cluster in time and space: tweets go viral, flurries of nearby earthquakes occur in quick succession. Temporal and spatial proximity are major factors in the chain reaction of events. However, how and why these clusters form is much more complex: the mechanisms that generate clusters of events are often self-exciting.
Models (e.g., Hawkes processes) that allow us to account for the inherent self-excitement without demanding knowledge of the generating mechanisms are becoming hugely popular in modelling the data examples given above. However, beyond the simplest form of the model, we currently lack appropriate measures of fit for these models. This project will look at goodness-of-fit for a range of self-exciting point process models, using simulation-based approaches to better understand how different cross validation regimes etc. affects inference.
No familiarity with point processes is needed; however, strong R programming skills are required and familiarity with C++ would be advantageous. Good grades in STATS310 and STATS330 would also be an advantage.
Project co-supervised with Alec van Helsdingen.
