Every Prediction Market Strategy, Simulated 100 Million Times

Can you actually beat Polymarket? I tested every popular prediction market strategy on 100 million simulated traders, using real markets that have already resolved. The data: 5,496 real yes/no Polymarket markets with $15.8 billion of trading volume, resolved between 2023 and 2026. Each market has its real price 30 days, 7 days and 1 day before it closed, and its real outcome. The markets come from two independent public research samples. On the markets both samples contain, their outcomes agree completely. How a bet works: A Yes share pays $1 if the event happens, and a No share pays $1 if it does not. Every bet buys at the market’s real recorded price, plus one price tick. From March 30, 2026 it also pays Polymarket’s taker fee for that category. Bets are held until the market resolves. Each simulated trader starts with $1,000 and makes 50 bets of $10, on real markets that fit its strategy. The strategies: • Buying Yes on any market • Always buying No, or picking Yes or No at random • Longshots under 10 cents and lottery tickets under 3 cents • Favorites over 50 cents, and underdogs under 50 cents • “Bonds” at 90–95 cents, and near-certainties at 95 cents or more • Toss-up markets priced 40–60 cents • Momentum and contrarian: backing the side that rose or fell • Swing trading: buying a week out and selling the day before • The last-day favorite, and the early-bird favorite a month out • Sports favorites only • The quant: a model fitted on earlier markets, betting only when it sees an edge • The martingale: doubling the stake after every loss • All in on 95-cent favorites, every single bet • A control: a copy of the first strategy on different random draws, to show the simulation is fair 20 strategies × 5,000,000 traders = 100,000,000 simulated traders. Every result also shows its real range of uncertainty, because all the traders draw from the same set of real markets. Extra tests outside the 100 million: no costs at all, no fees, a flat 1-cent spread, two price ticks, and Kalshi’s fee schedule. Each strategy is also tested on earlier and later markets separately. Checks: a separate program replays the bets and the fee rules and matches the simulator. The quant’s model never sees the markets it bets on. Limits: About 5,500 markets. The recorded prices are last trades, not guaranteed fills. Only markets with at least $20,000–50,000 of volume. Polymarket only, 2023–2026. The fee schedule is as reported. No taxes. Sources: github.com/CH4RL3I/honest-odds (MIT) and github.com/carol-747/polymarket-calibration, both built from Polymarket’s public data. Bürgi, Deng and Whelan, “Makers and Takers: The Economics of the Kalshi Prediction Market” (2025). DISCLAIMER This video is for entertainment and educational purposes only. It is not financial, investment, trading, betting or legal advice, and nothing in it is a recommendation to trade or bet on any market. All results come from a simulation using historical prices; real trading involves risks the simulation does not capture, and past results do not predict future results. You can lose some or all of the money you put in. Prediction markets may not be legal or available where you live, and you must be an adult to use them. Never bet money you cannot afford to lose. If gambling is causing you or someone you know problems, help is free and confidential: in the US call 1-800-GAMBLER (National Council on Problem Gambling); in the UK call the National Gambling Helpline on 0808 8020 133 (GamCare).