← Back to Trading and Market Microstructure
q-fin.TR

Can AI Traders Beat Crypto's Wild Swings with Pair Trading?

Damian Lebiedź, Robert Ślepaczuk

June 3, 2026

Pair trading—betting that two correlated assets will realign—works in stocks but breaks down in crypto's extreme volatility. The authors built an AI system using reinforcement learning (PPO with memory layers) to decide when to enter and exit trades within strict risk guardrails. Testing on Binance hourly data showed the AI beat heuristic baselines with 10% statistical significance, suggesting AI execution can stabilize the strategy where rigid rules fail—though the result falls short of 5% confidence, a reminder that crypto's chaos still defeats most trading schemes.
Published as Dynamic Multi-Pair Trading Strategy in Cryptocurrency Markets with Deep Reinforcement Learning arXiv:2606.04574
Read the original paper →