Predictive Methods for Hawkes Processes in High Frequency Finance

Abstract

High frequency financial data is burdened by a near obligatory level of randomness that obfuscates the task of predictive modelling. Given that events in a limit order book can be self-exciting in nature and influenced by many external sources, Hawkes processes are a common choice for modelling limit order book dynamics. Many stochastic models have been proposed to describe empirical order-book dynamics and to build a framework for prediction tasks. In this thesis an evaluation of one such family of models, the General Compound Hawkes Process models, is presented with the goal of forming a basis for practical applications of the model. Also examined is the broader application of Hawkes processes in developing trading signals that consolidate information contained within the order flow of a limit order book. Through this work we show the feasibility of Hawkes process based models by showing that they can be used to create both prognostic short term indicators and systematic long term predictions for the mid-price.