Calendar
Isabella Deutsch: Tweets and Apparel Wholesale: Bayesian Estimation of Hawkes Processes with Excitation and Inhibition
Abstract: Hawkes Processes provide a flexible way to model point process data that occurs in clusters or bursts. In this talk we take a close look at possible model specifications and their Bayesian estimation procedures. First, we examine a base case allowing for excitation with one data set, as showcased by a Twitter data example. Second, we sketch out a model that allows for self-influence and cross-influence with multiple data streams, including both excitation and inhibition. We highlight challenges, possible network and causal interpretation, and discuss real life applications using apparel wholesale data.