Logic-Q: Improving Deep Reinforcement Learning-based Quantitative Trading via Program Sketch-based Tuning

• DOI: 10.1609/aaai.v39i17.34045
• Publication Date: 2025-04-11T12:50:36Z
• Authors: Zhiming Li; Junzhe Jiang; Yushi Cao; Aixin Cui; Bozhi Wu; Bo Li; Yang Liu; Danny Dongning Sun
• Keywords: Sketch

Deep reinforcement learning (DRL) has revolutionized quantitative trading (Q-trading) by achieving decent performance without significant human expert knowledge. Despite its achievements, we observe that the current state-of-the-art DRL models are still ineffective in identifying market trends, causing them to miss good opportunity or suffer from large drawdowns when encountering crashes. To address this limitation, a natural approach is incorporate knowledge trends. Whereas, such abstract and hard be quantified. In order effectively leverage knowledge, paper, propose universal logic-guided deep framework for Q-trading, called Logic-Q. particular, Logic-Q adopts program synthesis sketching paradigm introduces model design leverages lightweight, plug-and-play trend-aware sketch determine trend correspondingly adjusts policy post-hoc manner. Extensive evaluations of two popular tasks demonstrate can significantly improve previous strategies.