Heterogeneous Trading Agents Show Distinct Stress and Performance Patterns in Automated Markets
ORIGINAL / Agentic AI in Automated Markets
The study uses reinforcement learning to simulate three types of trading agents and finds significant differences in their market stress patterns, risk-adjusted performance, and impact on liquidity and volatility, supporting behavior-specific regulation.
01 ABSTRACT
The study models a rational arbitrageur, a manipulator, and a retail herd-follower as reinforcement learning systems trained in a common environment to examine their market impact. Pilot experiments on hourly cryptocurrency data show systematic differences among the three agent types in stress frequency and duration, and in rolling Sharpe ratios. The authors propose behavior-specific regulation rather than uniform limits on volatility or turnover.
02 KEY FINDINGS
- Herd-followers produce the most frequent but shortest stress episodes; manipulators produce the fewest but most persistent; rational agents show the most stable rolling Sharpe ratios.
- SHAP attributions link herd behavior to drawdown and cross-asset signals, manipulator behavior to short-term volatility and position state, and rational behavior to momentum.
- The study conducts a pilot using hourly cryptocurrency data for AAVE, ETC, and SUI to motivate the hypotheses.
AI GENERATED SUMMARY / DISCOVERED BY ARXIV Q-FIN