Deep Reinforcement Learning for High Frequency Trading under a Directional Changes Sampling Paradigm

Advances in deep reinforcement learning (DRL) techniques have given rise to numerous impressive technologies in recent years, including record-breaking automated Atari agents and ChatGPT. Deep reinforcement learning also has significant applications in high frequency trading (HFT), offering a means to exploit market inefficiencies at extremely high speeds. Traditional HFT methods often struggle with making intelligent, real-time decisions due to the vast volumes of data and the need for rapid processing.

However, DRL addresses these challenges by enabling algorithms to continuously learn and optimise trading strategies through interaction with the market environment, thus improving adaptability and performance in dynamic settings. Additionally, the directional changes (DC) paradigm enhances this approach by focusing on significant price movements instead of fixed-time intervals, providing a more efficient and meaningful representation of market dynamics.

By reducing noise and data redundancy, DC sampling allows for more accurate and efficient market analysis. Combining DRL with the DC paradigm creates a robust framework for HFT, leading to smarter, faster decision-making and a deeper understanding of market behaviour.

Empirical results demonstrate the effectiveness of this integrated approach, showcasing its potential to transform high frequency trading by significantly enhancing trading efficiency and profitability.

Hosted by the Centre for Computational Finance and Economic Agents.