• optimal execution
• reinforcement learning
• trading environment

This thesis presents a reinforcement learning (RL) extension of the Almgren–Chriss model for optimal trade execution in dynamic market environments. A realistic and stochastic trading simulator is developed, modeling five key market variables (volatility, liquidity, bid-ask spread, market sentiment, and macroeconomic events) as dynamic processes influenced by agent behavior. A Proximal Policy Optimization (PPO) agent is trained to adaptively execute large orders under changing market regimes, aiming to minimize both expected costs and risk. A comprehensive sensitivity analysis evaluates the agent’s performance against the analytic benchmark across various market conditions. Results show that the RL agent learns robust, adaptive strategies and consistently outperforms the classical solution in volatile and complex environments.