Publication

Abstract :

In our article, we introduce an innovative method for improving stock portfolios through Reinforcement Learning(RL), specifically utilizing the proximal policy optimization (PPO) algorithm. Conventional portfolio optimization techniques often face challenges in adapting to the ever-changing and extensive financial markets, primarily concentrating on maximizing returns while lacking adequate risk management. To overcome these limitations, we suggest an RL-based framework where funds are allocated dynamically within portfolios, maintaining a balance between risk and return through a custom reward function that includes metrics such as the Sharpe ratio. The model was trained on historical stock data up to 2023 and evaluated with unseen data from 2024, demonstrating exceptional performance. These findings indicate the capabilities of models that exceed traditional benchmarks, including the S&P 500 and the equilibrium portfolio. Moreover, we have developed a real-time adaptability framework, making it well-suited for live trading scenarios. Our study emphasizes the promise of RL in contemporary portfolio management and offers investors a strong and adaptable approach for optimizing risk-adjusted returns in fluctuating financial markets.