Deep Learning-Driven Order Execution Strategies in High-Frequency Trading: An Empirical Study on Enhancing Market Efficiency

This paper analyzes the efficacy of deep learning models, in particular reinforcement learning, in optimising the order execution for HFT. Typical execution models such as volume-weighted average price (VWAP) and time-weighted average price (TWAP) are typically unsuitable for HFT as they lack flexibility and are also latency. To solve these problems, our work utilizes a PPO reinforcement learning algorithm. Based on high-frequency tick data, the model continuously adapts to market conditions and enables performance enhancements in execution quality, speed and versatility under volatile markets. The experiment showed that the deep learning model performed better than the VWAP baseline, filling at 92%, market impact reduced by 15%, and running 43% faster. As this work shows, adaptive, deep learning based strategies not only enhance HFT execution, but also promote stability in the market, highlighting the possibilities of AI as a practical tool for trading.