Turbulence-driven Autonomous Stock Trading using Deep Reinforcement Learning

This paper explores Deep Reinforcement Learning (DRL) algorithms for autonomous stock trading, aiming to replace stockbrokers with more efficient and profitable strategies. The study focuses on stock trading in finance, using DRL to analyze trends and exploit market fluctuations. It delves into the application of DRL techniques in stock trading, using Proximal Policy Optimization (PPO) algorithm, and, experimental results from training and testing on Dow 30 and S&P 500 datasets that reveal the effectiveness of incorporating turbulence indicators such as McClellan Oscillator (MCO) and Hindenburg principle. The incorporation of modified indicators: MCO and Hindenburg Omen, significantly influenced the trading system's performance. In comparing trading strategies, the turbulence-incorporated Proximal Policy Optimization (PPO-T) agent, initially funded with$1M exhibited consistent improvement and greater resilience, with slight portfolio value dips compared to the non-turbulence counterpart (PPO). Notably, PPO-T outperformed PPO during market turbulence, achieving portfolio value surged to 1.8 and 2.0 for DOW 30 and SP 500, respectively, surpassing the values of 1.5 and 1.9 attained by PPO. This underscores PPO-T's capacity for higher final portfolio values, emphasizing the efficacy of turbulence indicators in fortifying trading systems, especially in turbulent market conditions. The agent's performance evaluation involves tracking portfolio value changes over specified trading days.