Interactions of Market Making Algorithms: A Study on Perceived

Abstract

The widespread use of market making algorithms and the associated feedback effects may have unexpected consequences which need to be better understood. In particular the phenomenon of 'tacit collusion' in which the interaction of algorithms leads to an outcome similar to a collusion among market makers, has increasingly received regulatory scrutiny. We propose a game-theoretic model of a financial market in which multiple market makers compete for market share and learn from market data to adjust their spreads. We model this learning process through a decentralized multi-agent reinforcement learning algorithm and show that, even in absence of price information sharing, under specific mechanism through which market makers compete for market shares, market prices may converge to levels which are similar to a collusion situation, resulting in 'tacit collusion'. We briefly discuss implications of our research for market regulators.