Abstract

Identifiability can be defined as the existence of a unique maximum likelihood estimator for any arbitrary data set. When models become unidentifiable, issues occur as the fitted model can diverge significantly from the true data generating process, rendering inference and prediction essentially useless. Understanding when Hawkes processes become unidentifiable has yet to be fully explored; however, in much of the current literature identifiability is either assumed or ignored entirely. This paper discusses under what circumstances Hawkes processes become unidentifiable: providing a theoretical interpretation of how the structure of Hawkes processes affect the identifiability of the process itself.

Andrew Connell

Graduate Mathematician & Statistician

My research interests include signal processing, time series analysis and point processes.