CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: State-dependency and Hawkes processes for order book modeling Authors:  Ioane Muni Toke - CentraleSupelec (France) [presenting]
Sfendourakis Emmanouil - CentraleSupelec (France)
Abstract: The modeling of high-frequency occurrences of events in electronic limit order books has been for the past few years an active subject of research for both academic and practitioners, with potential applications in the understanding of market microstructure, execution problems, trading strategies development, market regulation, etc. Several contributions in this field use self-exciting Hawkes processes to account for the clustering of events. Recently, a state-dependent Hawkes process has been proposed to model high-frequency financial data, in which the excitation kernel is state-dependent. We investigate an alternative extension of Hawkes processes, in which state-dependency is added to the process intensity via a multiplicative term. More precisely, the multiplicative term is an exponential of a linear combination of observed state covariates, such as spread or the imbalance. From a practical point of view, this alternative definition allows using multiple exponential kernels without a cumbersome explosion of the parameter space dimension. It can be applied to all series of financial events, without having to track all-state modifications. We present empirical results for model selection as well as event type prediction for 30+ stocks traded on the French market in 2015.