High-Frequency Trading on Polymarket: A Cluster-Based Analysis of Trading Behavior

Sammanfattning

Polymarket, a decentralized blockchain-based prediction market, has been described as a "bot playground" where algorithmic traders shape market activity. Because the platform does not label trade behavior, the prevalence and role of high-frequency trading (HFT) remain difficult to assess. This thesis identifies and analyzes HFT in Premier League moneyline markets on Polymarket, using the 15 highest-volume matches from the 2025/2026 season. Features recording trade frequency, execution speed, and same-second activity were used in unsupervised clustering, combined with a rule-based classifier grounded in financial literature, to isolate wallets exhibiting HFT behavior. The results show that HFT wallets form a clear minority, with a bootstrap mean of 7.35\% (95\% CI: [6.84\%, 7.88\%]) of unique wallets per match, yet represent a bootstrap mean of 85.77\% (95\% CI: [82.58\%, 88.83\%]) of trading volume per match. No significant difference in median profit margin was found between HFT and non-HFT wallets, but the groups differed in risk profile. HFT wallets showed small, consistent margins, while non-HFT wallets showed far greater variation. These findings indicate that a small set of behaviorally distinct wallets dominates capital flow on these markets, raising questions about the integrity and structure of decentralized prediction markets.