Reading the result
The headline numbers point in different directions, and both are worth taking at face value for what they describe. On Kalshi, 10,061 settled whale buys won 77.3 percent of the time and returned 24.1 percent on stake gross: $242,597 on $1,006,100 staked. On Polymarket, 124,644 settled whale buys won 61.0 percent of the time and returned minus 0.17 percent on stake: the pooled whale trade on Polymarket in this window was, before costs, a coin flip with a very large sample. That gap is the difference between a win rate and a return. A win at a high price pays little, and Polymarket whales in this window were paying up.
Trades per market is the number to hold onto. Kalshi recorded about 1.8 whale trades per market against 14.4 on Polymarket. Ten buys in one market are ten rows and one outcome, so the Polymarket sample has far fewer independent bets than its 162,337 rows suggest, in either direction. Its 22,867 unresolved buys, 15.5 percent of the eligible set, also show how recent the tail is: a window ending the day before export has many open positions, and the settled sample leans toward markets that resolve quickly, such as sports and short-dated crypto.
None of this is a recommendation to copy the average whale on either venue, and the two pipelines do not record the same population: Kalshi identity is only available where a trader shares a public name, and its whale floor differs from Polymarket's. The useful conclusion is narrower. Whale size alone is not a strategy on Polymarket, and the edge, where it exists, lives in subsets: a price band, a category, a trader with a graded record. That is what the backtester and ranked traders are for, and the events behind this study are the ones they run on.
Population and method
- Window. Detection time from 2026-08-14T00:00:00.000Z inclusive to 2026-09-13T00:00:00.000Z exclusive.
- Lane. Whale lane only: trades above each venue's size floor, whoever placed them. Tracked-lane trades by identified traders below the floor are excluded.
- Eligibility. buy_yes and buy_no with a recorded price strictly between 0 and 1. Invalid prices are counted and excluded.
- Scoring. Resolution observed at export time: win, loss, void or unresolved. Only wins and losses enter the rate and the PnL.
- Stake. $100 per eligible settled buy, filled at the recorded price, held to settlement. PnL is 100 / price − 100 on a win and −100 on a loss.
- Not modelled. Fees, spread, slippage, entry delay, partial fills, depth, position limits, capital lock-up, and correlation between trades in the same market.
- Categories. Reported as unknown because labels were not verified for research use at export. No category claim is made.
- Export. One read-only, repeatable-read transaction against the production history, so every count comes from the same instant.
The full definitions, including the settlement rule and the payoff arithmetic, are on the methodology page.