Agent-Based Model Exploration of Latency Arbitrage in Fragmented Financial Markets

Computerisation of the financial markets has precipitated an arms-race for ever-faster trading. In combination, regulatory reform to encourage competition has resulted in market fragmentation, such that a single financial instrument can now be traded across multiple venues. This has led to the proliferation of high-frequency trading (LIFT), and the ability to engage in latency arbitrage (taking advantage of accessing and acting upon price information before it is received by others). The impact of LIFT and the consequences of latency arbitrage is a contentious issue. In 2013, Wah and Wellman used an agent-based model to study latency arbitrage in a fragmented market. They showed: (a) market efficiency is negatively affected by the actions of a latency arbitrageur; and (b) introducing a discrete-time call auction (DCA) eliminates latency arbitrage opportunities and improves efficiency. Here, we explore and extend Wah and NVellman's model, and demonstrate that results are sensitive to the bid-shading parameter used for zero-intelligence (ZIC) trading agents. To overcome this, we introduce the more realistic, minimally intelligent trading algorithm, ZIP. Using ZIP, we reach contrary conclusions: (a) fragmented markets benefit from latency arbitrage; and (b) DCAs do not improve efficiency. We present these results as evidence that the debate on latency arbitrage in financial markets is far from definitively settled, and suggest that ABM simulation-a form of decentralised collective computational intelligence-is a productive method for understanding and engineering financial systems.