Dynamic Profits: Leveraging Reinforcement Learning in Evolving Financial Markets

Abstract-Electronic trading or algorithmic trading has changed the landscape of financial markets as data is processed and analyzed, which enables instant decision making. The application of reinforcement learning (RL) in algorithmic trading has the ability of constant improvement and optimization in ever- changing environments. Autonomous, intelligent systems that can operate in the unpredictable financial market conditions are required at an ever-growing rate. Trading agents can learn market optimal decision-making strategies through reinforcement learning, which makes it a good fit for real-time usage. The primary aim of this study is to overcome the challenge posed by the traditional algorithmic trading approaches that target high market volatility and non-stationary data using pre- programmed strategies. Most of the published studies are concentrated on the theoretical aspects of the models while very little attention is given to their application, transaction cost, slippage, and market impact. In RL based trading systems, learning needs to be stable or the trader risks overfitting, setting risk parameters for exploration and exploitation can also Markov Decision Process be very difficult. We develop a custom RL framework that compensates for transaction costs at the breakeven point, where other methods fail. Rather than focusing on other reward functions, our method can actually be implemented in real-time trading situations.