Bandit Market Makers

We propose a flexible framework for profit-seeking market-making, using a sequence of cost-function based automated market-makers with bandit learning algorithms. We do this by considering the magnitude to which a cost-function extends beyond the simplex as a bandit arm, and the minimum-expected profits consistent with a no-arbitrage condition as the rewards. This allows for the creation of market-makers that can adjust bid-asks spreads dynamically, maximising worst-case-expected profits.