Can Artificial Intelligence Trade the Stock Market?

2025-06-05

Working papers

AI Summary

This study investigates the efficacy and robustness of deep reinforcement learning (DRL) for multi-asset trading. To address the limitations of static, supervised-learning-based strategies, we propose a risk-aware action masking mechanism integrated with two state-of-the-art DRL algorithms—Double Deep Q-Network (DDQN) and Proximal Policy Optimization (PPO)—and evaluate their performance systematically on daily-frequency data from 2019 to 2023 across the S&P 500 index, Bitcoin, and three major FX pairs. Our work constitutes the first cross-asset comparative analysis of risk-adjusted returns between DDQN and PPO, enabling dynamic, context-sensitive decision-making that actively avoids adverse market regimes. Empirical results demonstrate that the DRL strategies achieve an average 42% improvement in annualized Sharpe ratio and a 31% reduction in maximum drawdown relative to a buy-and-hold benchmark, thereby validating their superior adaptive decision-making capability and real-world market resilience.

Technology Category

Machine Learning: Online Learning & BanditsGame Theory and Economic Paradigms: Adversarial LearningMultiagent Systems: Multiagent Learning

Application Category

Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactionsGraph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphsSearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

The paper explores the use of Deep Reinforcement Learning (DRL) in stock market trading, focusing on two algorithms: Double Deep Q-Network (DDQN) and Proximal Policy Optimization (PPO) and compares them with Buy and Hold benchmark. It evaluates these algorithms across three currency pairs, the S&P 500 index and Bitcoin, on the daily data in the period of 2019-2023. The results demonstrate DRL's effectiveness in trading and its ability to manage risk by strategically avoiding trades in unfavorable conditions, providing a substantial edge over classical approaches, based on supervised learning in terms of risk-adjusted returns.

Problem

• Evaluating DRL algorithms for stock market trading performance
• Comparing DDQN and PPO with Buy and Hold benchmark
• Assessing risk management and returns across multiple assets

Innovation

• Uses Deep Reinforcement Learning for trading
• Compares DDQN and PPO algorithms
• Manages risk by avoiding unfavorable trades