Design and Implementation of a Reinforcement Learning Agent for Algorithmic Trading Simulation

Financial markets are characterized by high volatility, non-linearity, and complex dynamics, rendering the design of manual trading strategies both difficult and prone to human bias. While supervised learning models have been employed for price prediction, they often fail to capture the sequential nature of trading decisions. This paper presents a simulation framework for an automated stock trading bot utilizing Reinforcement Learning (RL). Specifically, a Deep Q-Network (DQN) agent is implemented and trained to make discrete trading decisions (Buy, Sell, Hold) based on historical price data and technical indicators. The system utilizes a custom OpenAI Gym-compatible environment to simulate market conditions and employs reward shaping to optimize for risk-adjusted returns. Experimental results demonstrate that the RL agent outperforms traditional rule-based strategies, such as Buy-and-Hold and Moving Average Crossover, in terms of cumulative return and Sharpe ratio within the simulated environment.