Title: Modelling limit order book data by state-dependent Hawkes processes
Authors: Mikko Pakkanen - Imperial College London (United Kingdom) [presenting]
Maxime Morariu-Patrichi - Imperial College London (United Kingdom)
Abstract: During the past ten years, self-exciting Hawkes processes have become a popular model for high-frequency financial data, as they are able to capture the endogeneity and feedback effects in order flow data at very short time scales. In such a model based on Hawkes processes, the arrival rate of new orders depends on the past order flow, but it cannot directly depend on any state variables of the limit order book, such as the current bid/ask price or queue imbalance. To address this limitation, we introduce a novel state-dependent extension of a Hawkes process. The new framework couples the Hawkes process to a state process that influences the arrival rate of new orders whilst the arriving orders may, reciprocally, prompt the state process to move to a new state. We develop maximum likelihood estimation methodology for the new class of processes and apply it to NASDAQ limit order book data. The empirical results indicate that excitation effects in order flow depend strongly on queue imbalance.