Effect of Market Spread Over Reinforcement Learning Based Market Maker

Market Making (also known as liquidity providing service) is a well-known trading problem studied in multiple disciplines including Finance, Economics and Artificial Intelligence. This paper examines the impact of Market Spread over the market maker’s (or liquidity provider’s) convergence ability through testing the hypothesis that “Knowledge of market spread while learning leads to faster convergence to an optimal and less volatile market making policy”. Reinforcement Learning was used to mimic the behaviour of a liquidity provider with Limit Order Book using historical Trade and Quote data of five equities, as the trading environment. An empirical study of results obtained from experiments (comparing our reward function with benchmark) shows significant improvement in the magnitude of returns obtained by a market maker with knowledge of market spread compared to a market maker without such knowledge, which proves our stated hypothesis.