Defy the Game: Automated Market Making using Deep Reinforcement Learning

Abstract

Automated market makers have gained popularity in the financial market for their ability to provide liquidity without needing a centralized intermediary (market maker). However, they suffer from the problems of slippage and impermanent loss, which can lead to losses for both liquidity providers and takers. This work implements a pseudo-arbitrage rule to solve the impermanent loss issues related to arbitrage opportunities. The mechanism implements a trusted external oracle to get the market conditions, put them on the automated market maker, and match the bonding curve to them. Next, the application of a Double Deep Q-Learning reinforcement learning algorithm is proposed to reduce these issues in automated market makers. The algorithm adjusts the curvature of the bonding curve function to adapt to market conditions quickly. This work describes the model, the simulation environment used to learn and test the proposed approach, and the metrics used to evaluate its performance. Finally, it explains the results of the experiments and analysis of their implications. The approach shows promise in reducing slippage and impermanent loss and recommending improvements and future works.