🤖 AI Summary
This work addresses the inefficiencies in decentralized prediction markets—specifically, inadequate information aggregation, mispricing, and transaction latency—that collectively erode arbitrage opportunities. To tackle these challenges, the paper introduces the first prediction framework based on a multi-agent large language model (LLM) architecture, integrating 50 diverse LLM agents. The system employs confidence-weighted Bayesian fusion to combine market-implied probabilities with crowd consensus, executes trades under risk constraints using a quarter-Kelly criterion, and identifies cross-market pricing inefficiencies via Kullback–Leibler and Jensen–Shannon divergences. Furthermore, it leverages a log-normal model to infer centralized exchange prices, enabling latency arbitrage within human reaction-time windows. Empirical results on Polymarket demonstrate substantially improved probability calibration, with Brier scores and log-loss approaching those of human superforecasters and significantly outperforming single-model baselines.
📝 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. PolySwarm deploys a swarm of 50 diverse LLM personas that concurrently evaluate binary outcome markets, aggregating individual probability estimates through confidence-weighted Bayesian combination of swarm consensus with market-implied probabilities, and applying quarter-Kelly position sizing for risk-controlled execution. The system incorporates an information-theoretic market analysis engine using Kullback-Leibler (KL) divergence and Jensen-Shannon (JS) divergence to detect cross-market inefficiencies and negation pair mispricings. A latency arbitrage module exploits stale Polymarket prices by deriving CEX-implied probabilities from a log-normal pricing model and executing trades within the human reaction-time window. We provide a full architectural description, implementation details, and evaluation methodology using Brier scores, calibration analysis, and log-loss metrics benchmarked against human superforecaster performance. We further discuss open challenges including hallucination in agent pools, computational cost at scale, regulatory exposure, and feedback-loop risk, and outline five priority directions for future research. Experimental results demonstrate that swarm aggregation consistently outperforms single-model baselines in probability calibration on Polymarket prediction tasks.