A Stock Trading Strategy Based on Deep Reinforcement Learning from Multiple Price Charts

In recent years, machine learning techniques have been predominantly applied for automated financial trading. These techniques, however, often fall short because of the complexity and dynamic nature of financial markets. These markets are influenced by factors such as macroeconomic indicators and market sentiment, which simple models struggle to capture. Additionally, financial data are highly noisy and nonstationary, complicating the task of distinguishing meaningful signals from random fluctuations. Given the complexity of financial markets and the long-term nature of desired results, alternative approaches such as reinforcement learning (RL) appear to offer a more effective solution to adapt to changing market conditions and develop complex trading strategies over time, aiming at long-term benefits/profits. This adaptability is crucial in the volatile and multifaceted world of financial markets, where traditional models may fail to swiftly adjust. In addition, combining RL with multiple graphical representation price charts may significantly enhance financial trading by enabling algorithms to adapt dynamically across varying market conditions with a comprehensive, multitime frame perspective. Using multiple price charts provides the method with context on both immediate price movements and broader trends, helping it differentiate between noise and valuable signals. Accordingly, this article proposes a deep RL model that integrates various graphical price chart types. Experiments conducted on the S&P 500 index demonstrated higher profitability and lower risk with strategies utilizing Cartesian charts, polar charts, and Japanese candlesticks, compared to those using single price chart types. Furthermore, at the end of a two-year testing period, our multistrategy approach outperformed single trading approaches in 50 out of 52 financial evaluations.