Polybot

Under 0.04% of wallets took more than 70% of the profit.

Where the money goes

Prediction markets are brutally lopsided. A December 2025 on-chain analysis of 1.7 million Polymarket wallets found that about 70% lost money, while fewer than 0.04% of addresses captured more than 70% of all realised profit (DeFi Oasis, 2025). A second study in April 2026, covering 2.5 million wallets, put the share of losers at 84.1%. Only 2% of traders ever cleared $1,000, and just 840 addresses, 0.033%, made more than $100,000 (Shriber, 2026; Capozzalo, 2026). A June 2026 paper from the Centre for Economic Policy Research, covering 588 million trades, found that the top 1% of profitable users captured 76.5% of all profits (Akey et al., 2026). Most of these figures count wallet addresses, not people, so the true concentration among individuals is somewhat different.

That lopsidedness is the opportunity. If the profit sits with a few traders, the useful question is whether you can tell who they are in advance rather than after the fact. Some of those top accounts are bots and market makers. The ones I was after trade on better information than everyone else, whether that is inside knowledge or simply better research. The CEPR authors think insider trading is unlikely to explain the largest winners, which is a fair reminder that informed and inside are not the same thing.

How the system picks them

The scorer works in two stages and only ever looks backwards. Each quarter it ranks every trader on their last two years of results and keeps a shortlist of around ten thousand. Every day it re-scores that shortlist on the past 30 days and follows the top 200. A trader earns a place by being right consistently, and by buying into markets that were mispriced when they bought, not by winning a lot of safe bets at expensive prices.

Everything is tested walk-forward. The system picks traders using only data up to a date, then I watch what they do over the following six months. Nothing in the test can see the future. Here is what came out.

Do they keep making money?

This is the question that matters most, because a lucky streak looks the same as skill until it stops. The chart shows the return on every bet the picked traders made, before trading costs, on two separate windows. The second, the first quarter of 2025, was never used to build the system. The result barely moves: +10.1% on the original window and +9.7% on the unseen one. A fluke rarely repeats that closely. Skill does.

What did they actually earn?

Over the same six months, a $10,000 wallet placing the same bets as the picked traders grew by 161.5%, before any trading costs. That figure measures what these traders earned. It is not a promise of what someone following them would make, because following costs money: with a flat 3% taken off every bet the same wallet returns 60.3%. For scale, the S&P 500 rose 18.7% and the FTSE 100 8.3% between the same dates (S&P Dow Jones Indices, 2026; Yahoo Finance, 2026), both as price returns. The worst peak-to-trough fall along the way, the maximum drawdown, was 7.8% before costs and 10.0% after slippage. I also ran the same bot on randomly chosen traders ten times. It lost money every time, so the gain comes from who gets picked and not from the bot.

Are they just backing favourites?

The edge shows up at every price. Each bar is the gap between how often the picked traders won and how often the market's price said they would. Anything above zero means they beat the odds they were given. They do it on long shots, on favourites and everywhere between, so they aren't simply buying the obvious winner. The random traders did the opposite: they won more bets, 63% against 44%, and still lost money because they paid too much for them.

Is it one lucky wallet?

The edge is strongest when the system follows about 200 traders. Narrow it to the top ten and it gets worse, and most of the profit then comes from a single account. The skill is spread through a crowd of traders rather than sitting in a handful of names, which is what makes the method sturdier than a stock tip.

What's left to solve

Following them in real time is the hard part. By the time a copy order reaches the market the price has often moved, and many orders never fill, which takes a good share of the edge. That's the next job, and the project report covers it in detail. Picking the traders was the part I wanted to prove first, and it works.

Data

151 million trades and 1.9 million resolved markets, Nov 2022 to Oct 2025

Built with

Python and Polars, 23 numbered analysis scripts

Status

Research only. No real money traded, and not financial advice. Figures are from the June 2026 report.

References

1. Akey, P., Vincent, G., Harvie, N. and Martineau, C. (2026) 'Who wins and who loses in prediction markets? Evidence from Polymarket', CEPR Discussion Paper No. 21615. London: Centre for Economic Policy Research. Available at: https://cepr.org/publications/dp21615 (Accessed: 9 October 2026).
2. Capozzalo, O. (2026) '84% of Polymarket traders are losing money, new research finds', The Defiant. Available at: https://thedefiant.io/news/research-and-opinion/polymarket-profitability-report-april-2026 (Accessed: 9 October 2026).
3. DeFi Oasis (2025) Analysis of realised profit and loss across 1.7 million Polymarket addresses [X post], late December. Available at: https://twitter.com/defioasis/status/2005591714057297973 (Accessed: 9 October 2026). Figures read via Le, V.T. (2025) 'Polymarket's winner-take-most reality emerges as 70% of traders lose money', Coin360, 30 December. Available at: https://coin360.com/news/polymarket-traders-losses-profit-concentration-2026-predictions (Accessed: 9 October 2026).
4. S&P Dow Jones Indices LLC (2026) S&P 500 [SP500], retrieved from FRED, Federal Reserve Bank of St. Louis. Available at: https://fred.stlouisfed.org/series/SP500 (Accessed: 9 October 2026).
5. Shriber, T. (2026) 'New study suggests 84% of Polymarket traders aren't profitable', Casino.org, 7 April. Reports the April 2026 analysis by A. Sergeenkov. Available at: https://www.casino.org/news/84-of-polymarket-traders-arent-profitable/ (Accessed: 9 October 2026).
6. Yahoo Finance (2026) FTSE 100 (^FTSE) historical prices. Available at: https://finance.yahoo.com/quote/%5EFTSE/history/ (Accessed: 9 October 2026).

Every other figure on this page is from my own backtests and the project report. I read the studies above through news coverage and haven't checked them against the original datasets.