A Market Impact Model for Limit Order Book Simulators

Abstract

The purpose of this thesis is to develop a market impact model that can be used to penalize the testing of trade execution strategies in simulation environments. Traditional simulators are used primarily by algorithmic brokers to determine optimal execution strategies, particularly for trades of large sizes, that are efficient and minimize transaction costs for clients. However, these simulators do not incorporate market impact due to both the complexity of determining how every trade affects the price of the traded asset and the need to maintain the similarity between the simulated market and historical, real market data. We implement a VAR model for market impact, calibrated on futures data, that is representative of an opponent trader and can be used separately from the simulator to apply penalties to executed trade prices and more accurately quantify costs. We optimize against this model to design an execution strategy that sends orders at regular, medium-length trade intervals to minimize predicted impact. We formulate this strategy by analyzing trade flows over time periods of different lengths to predict the optimal trade interval length. We test this strategy against the standard strategy of sending orders at regular time intervals and we discover that the trade interval strategy incurs lower costs, with our predicted trade interval of 78 trades as the optimum. Considering that market impact models have not been widely used with simulators for algorithmic trading, our methodology and results provide meaningful contributions to existing trade execution literature and lay the groundwork for future research into improving trade execution with the use of predictive impact models.