Flash Boys or Superforecasters? Prediction Market User Base
Ethereum Korea One: Genesis · Day 2 · 2026-09-29 · 24:38 · In English
Who is really driving prediction market volume: skilled forecasters, retail bettors, or bots? At Ethereum Korea One: Genesis, Pantera Capital's Danning Sui breaks down the data behind Polymarket and Kalshi.
Speakers
- Danning Sui, Research Director, Pantera Capital
About this session
Danning Sui, Research Director at Pantera Capital, previously at Flashbots looking at DEX trading data and MEV, presents an analysis of prediction market data from the first half of this year covering Kalshi, Polymarket International, and Polymarket US. The talk covers the metrics that matter, how incentives shape volume, who wins and loses among users, and an MEV strategy found in short-term crypto markets.
Key takeaways
- A CFTC DCM license lets a platform define outcomes as binary contracts and operate across US states at the federal level, unlike state-by-state sports betting. Polymarket International matches orders offchain but settles on Polygon, which allows wallet-level analysis, while US regulated platforms are fully offchain, making wash trading harder to detect.
- Sui recommended looking at single-side taker volume, open interest, and turnover (volume divided by open interest) together, since notional volume is calculated differently on every platform. In July, Kalshi led with about $13 billion in monthly volume and about $1 billion in open interest, while Polymarket International and US combined did about $5 billion. In September, Polymarket US (about $2 billion in taker volume) flipped International (about $1.99 billion).
- Polymarket US is almost entirely sports, which gives it the highest turnover. At Polymarket International, 98% of fees come from crypto and sports, while geopolitics markets are intentionally kept at zero fees. Kalshi's crypto markets have little open interest but show six times daily turnover.
- Critics recently pointed to Kalshi's CFTC filings suggesting it rebates makers the same amount it charges in fees, effectively allowing zero-cost volume (Kalshi denied this, and the filing concerned its perp volume). Polymarket's crypto volume also fell after its maker rebate program changed. Sui's conclusion: incentivized volume is "not sticky."
- Wallets trading more than 100 times a day or in over 1,000 markets were classed as bots, and those putting $50,000 or more into one category as informed traders. Of about 2 million wallets active in the first half on Polymarket International, 89% behaved like retail but generated only about 20% of volume, while bots and pros (about 10% of wallets) generated about 77% of taker volume. Bots earned about $100 million a year as a group, yet the median bot wallet lost $70, which Sui linked to experimental AI agent bots built on the Polymarket API. Retail lost the most, and 146 wallets were profitable every month.
- In 5 and 15 minute BTC markets, 80% of bots trade against other bots. Near market close, Binance BTC volume can spike 12 times while the price briefly jumps and then reverts, suggesting position holders push the Chainlink-based closing price. BTC moves about 5bp on average over 5 minutes versus about 40bp over 4 hours, and 81% of 5 minute markets close within ±0.05% of the starting price, so short markets are cheap to manipulate, and Polymarket has moved toward a TWAP benchmark.
Chapters
- Introduction and agenda
- The prediction market landscape and regulation
- Polymarket US vs. International: data access and volume
- Metrics that matter: taker volume, open interest, turnover
- Market categories and fee capture
- Kalshi's crypto growth and the incentive debate
- User cohorts: retail, pros, and bots
- PnL: who wins and who loses
- MEV in short-term crypto markets
- Key takeaways