PolySwarm: A Multi-Agent Large Language Model Framework for Prediction Market Trading and Latency Arbitrage
Summary: arXiv:2604.03888v1 Announce Type: new
Abstract
This paper presents PolySwarm, a novel multi-agent large language model (LLM) framework designed for real-time prediction market trading and latency arbitrage on decentralized platforms such as Polymarket. The framework utilizes an innovative approach to enhance trading efficiency and accuracy in predicting binary outcomes.
Key Features of PolySwarm
- Multi-Agent System: PolySwarm deploys a swarm of 50 diverse LLM personas, each contributing to the evaluation of binary outcome markets.
- Probability Aggregation: The system aggregates individual probability estimates through a confidence-weighted Bayesian combination of swarm consensus with market-implied probabilities.
- Risk-Controlled Execution: The implementation applies quarter-Kelly position sizing to ensure risk-controlled execution of trades.
Information-Theoretic Market Analysis
One of the standout features of PolySwarm is its incorporation of an information-theoretic market analysis engine. This engine utilizes:
- Kullback-Leibler (KL) Divergence: This metric helps to quantify the divergence between two probability distributions, essential for identifying inefficiencies.
- Jensen-Shannon (JS) Divergence: This measure is used to assess the similarity between two probability distributions, aiding in the detection of mispricings across different markets.
Latency Arbitrage Module
PolySwarm includes a latency arbitrage module that exploits stale prices on Polymarket. This module derives Centralized Exchange (CEX)-implied probabilities from a log-normal pricing model and executes trades within the human reaction-time window, optimizing trading opportunities.
Evaluation Methodology
The paper provides a comprehensive evaluation methodology using Brier scores, calibration analysis, and log-loss metrics. These metrics are benchmarked against the performance of human superforecasters to establish the effectiveness of the PolySwarm framework.
Challenges and Future Directions
Despite its innovations, PolySwarm faces several challenges, including:
- Hallucination in agent pools, which may affect prediction accuracy.
- Computational costs at scale, as the system requires significant resources.
- Regulatory exposure, given the evolving landscape of decentralized finance.
- Feedback-loop risk, which may arise from the interaction between agents.
The authors outline five priority directions for future research to address these challenges and enhance the framework further.
Conclusion
Experimental results demonstrate that swarm aggregation consistently outperforms single-model baselines in probability calibration on Polymarket prediction tasks, indicating the potential of PolySwarm as a pioneering tool for decentralized prediction market trading.