Enhancing Risk-Adjusted Returns in Stock Trading: A Node2Vec-RL Approach

Abstract

This research introduces novel approach in the domain of stock prediction and trading through the advancement of reinforcement learning (RL) techniques. We present significant contributions across three pivotal areas: formulating stock market as graph, the integration of graph embeddings with RL agents, and the enhancement of interpretability within RL systems. This novel approach facilitates more sophisticated decision-making by the RL agent, resulting in enhanced pattern recognition and market trend analysis, thereby improving prediction accuracy and trading performance. Furthermore, we demonstrate the superior capabilities of the TD-3 algorithm in navigating the volatile and complex landscape of financial markets as compared to other advanced trading RL agents. Lastly, we address the critical aspect of interpretability in RL by implementing post-processing and visualization techniques to elucidate the decision-making process of the RL agent. This transparency not only builds trust in RL-based trading systems but also provides actionable insights, allowing practitioners to refine their strategies. Overall, our research significantly contributes to the fields of machine learning and financial trading, offering innovative tools and methodologies for enhancing the efficacy and understanding of RL applications in stock prediction and trading.