Prediction markets leak ≈ $40 M/yr in arbitrage Here’s the data 👇 s/o @OSGbyte

1️⃣ The dataset They scraped Polymarket (Apr ’24 → Apr ’25): 7 051 single-condition & 662 multi-outcome markets showed at least one window.

2️⃣ Two flavors of edge • Market-rebalancing = fixing bad odds inside one market • Combinatorial = exploiting logical links across markets See screenshot for breakdown by market category

3️⃣ Realized gains On-chain profit actually captured: $39.6 M: low-risk dollars left on the table.

4️⃣ Where the money was • Politics = biggest single trades • Sports = most frequent mis-pricings

5️⃣ Who exploited it A handful of bot wallets ruled: top 10 took > $8 M; one address alone ≈ $2 M.

6️⃣ Why inefficiency persists Polymarket’s hybrid CLOB isn’t atomic: fills can slip, so only spreads ≥ 5 ¢ are “risk-free.” See example in screenshot with the question "Will Assad remain President of Syria through 2024"

7️⃣ Fixes & design ideas - Clearer market-creation rules + batch auctions could tighten prices without killing liquidity. - cross platform arbitrage across multiple prediction markets

Bottom line Prediction markets are not perfectly efficient; vigilant traders (or bots) can still farm mispricings at scale. Next election cycle will be even juicier.