Polymarket Trading Strategies: Costs, Risks and Evidence

Last updated

Six strategies cover most of what traders do on Polymarket: making markets, arbitrage, trading on news, buying near-certain outcomes, buying long shots and copying skilled wallets. Each has a cost that can eat a thin edge: the taker fee, the spread, or a disputed market holding your money. This page goes through them with the numbers we could read and the studies we could check, and you can test the copying strategy on paper, free at Polymarket's prices with its fee taken out. It reports; it does not tell you what to do, and nothing here is a promise of a return.

For the wider picture, see how to make money on Polymarket, how Polymarket works and Polymarket vs Kalshi.

What is market making on Polymarket?

Market making means continuously posting bids and asks, and trying to earn the spread between them. Polymarket's documentation names GTC and GTD limit orders as the primary order types for passive market making. Orders cannot be edited in place, so changing a quote means cancelling it and posting a new one, and the documentation says inventory, the exposure you hold after fills, has to be part of every pricing and sizing decision.

The costs and the payments are both written down. Makers are never charged a fee; only takers pay. Polymarket's Maker Rebates pay makers a share of taker fees (15% to 25% by category, none in geopolitics or combos), and Liquidity Rewards pay for resting orders that sit inside a market's maximum spread. Neither is certain: Polymarket says the rebate percentage is at its sole discretion, and rewards under $1 are not paid.

On evidence, the cleanest study is of Kalshi, not Polymarket. Burgi, Deng and Whelan (George Washington University, working paper 2026-001, February 2026) found that on Kalshi makers earned an average return of -9.64% against -31.46% for takers, so makers lost less, not that they won. They add that makers must be ready to cancel their orders when new evidence suggests a bad price. The Yale and London Business School authors put market makers together with skilled takers at under 3.5% of accounts, taking over 30% of total gains on Polymarket.

Does arbitrage work on Polymarket?

Arbitrage means buying a set of outcomes whose prices do not add up. In an event where exactly one outcome resolves Yes, the prices of all the outcomes should add up to $1; if they cost less, buying them all locks in the difference, in theory. Saguillo and co-authors (arXiv 2508.03474, submitted 5 August 2025) describe two forms, rebalancing inside one market and combinatorial arbitrage across related markets, and estimate that about $40 million of profit was realised on Polymarket. That is the authors' estimate; we have not seen it replicated.

Polymarket's neg-risk design ties outcome prices together: a No share in one market can be converted into a Yes share in every other market of the event. The same documentation warns that the "Other" outcome in an augmented event can change its definition, so it advises against trading it directly.

Between venues, Gebele and Matthes (arXiv 2601.01706) found that equivalent markets show persistent gaps averaging 2 to 4%, and attribute them to structural frictions such as different event definitions. A gap that looks like free money can be two different questions with different resolution rules.

The costs are concrete. Each leg taken from the book pays the taker fee; at a 0.05 rate, 100 shares at 60¢ carry a fee of $1.20, and an arbitrage across several outcomes pays that kind of fee several times. Fill prices on a thin book can move between legs. The arithmetic only works if the gap is larger than all of that.

Can you trade on news?

Trading on news means buying or selling when information arrives, before the price has moved. The studies we read do not measure news trading directly. What they show is that few accounts trade with skill: the Yale and London Business School study, which covered two years of Polymarket trading, found about 3% of accounts do. Pew Research Center followed 11,989 active wallets in 2026 and found 58% gained or lost less than $100, and the average trade was $6.50.

There are mechanical costs too. Every taker trade in a market with a fee pays it, so news that moves the price by less than the fee and the spread leaves a trade that is right still losing money. In sports markets, Polymarket's documentation says open limit orders are cancelled automatically once the game begins, so a resting order does not carry through the start of play. Separately, Galaxy Research found about 4.1% of accounts automated (its cut-off: 50 orders per active day), placing 80.8% of orders on polymarket.com.

Is buying at 95–99¢ safe?

The arithmetic: a share bought at 95¢ pays $1.00 if it wins, so the profit is 5¢; if it loses you lose the 95¢. The break-even win rate before fees equals the price, 95%. The taker fee is small there: for 100 shares, $0.19 at 95¢ and $0.04 at 99¢ at a 0.04 rate, and $0.24 and $0.05 at 0.05. The price tells you how little room there is.

What the evidence says depends on the platform, so each figure is labelled. On Kalshi, Burgi, Deng and Whelan found that Kalshi contracts priced above 70¢ earned small but statistically significant positive returns after fees, and give the arithmetic: a Kalshi contract at 95¢ that wins 98% of the time returns +3.1% before fees. On Polymarket we have only our own data, for one week. In our report for week 41, dated 11 October 2026, covering positions settled by wallets in our directory between 4 and 11 October 2026, shares bought at 90¢ to 98¢ ended in profit 89.7% of the time and returned -0.9%; at 98¢ or more, 87.4% and -2.0%. One week of one data set says little about another week.

Then there is resolution risk. As Polymarket's documentation described it on 16 November 2026, its prediction markets resolve through UMA: someone proposes an outcome, there is a 2-hour challenge period, and if the outcome is disputed the process takes 4 to 6 days in total. Your money stays in the position while that happens. The outcome follows the market's written rules, not its title. A holder at 99¢ gains 1¢ when it goes right and loses the 99¢ when it goes wrong; in the rare 50/50 case each token redeems for $0.50, which costs 49¢ on a 99¢ share, and a dispute delays the money rather than costing it. You can copy whales' 95¢ to 99¢ buys on paper, with an entry-price filter, and see what the fee and the losses do to the average.

Do long shots pay?

The favourite-longshot bias is the pattern that cheap contracts win less often than their price implies. On Kalshi, Burgi, Deng and Whelan, in a sample of over 300,000 contracts, found that contracts costing 10¢ or less lose over 60% on average, and give a Kalshi example: a 5¢ contract that wins 3% of the time returns -40% before fees. They also report that takers lose more than makers, and that the bias holds across volume groups.

That is a Kalshi result, not a Polymarket one. In our own one-week sample on Polymarket (the week 41 report of 11 October 2026, positions settled 4 to 11 October), shares bought under 10¢ ended in profit 13.5% of the time and returned +18.4%, carried by a few long shots, while the 10¢ to 30¢ band returned -24.5%. The sign is the opposite of Kalshi's. One week, one data set and a handful of big winners can produce that, and it does not show that long shots pay on Polymarket in general.

Does copying whales work?

The case for it rests on a fact: trades and positions on Polymarket are recorded on-chain and publicly verifiable, so a wallet's history can be read and followed. The case against rests on what skill looks like. The Yale and London Business School study found that 44% of accounts classed as skilled in one half of events stayed skilled in the other half, against about 10% for active mutual funds in the authors' parallel test (Yale Insights; The Block). That is persistence, but not a majority. The Block's report of the paper adds that 12% of the top earners overlap with the skilled group and that about 60% of the "lucky winners" went back to losses on a separate sample.

Our own forward test is on the copy trading page. We picked wallets using only copies that settled before 1 September, then checked the same wallets on copies from September to 2 November, which played no part in the picking. It covered 1,036 wallets and about 92,000 copies, with a markup for slippage and the market's fee, held to resolution. In all three groups of wallets we show, the check result was negative: -2.9%, -6.7% and -10.4% a copy in non-sport markets, and -10.9%, -12.1% and -7.5% in sport. The conclusion there is that picking wallets by leaderboard rank did not work. It is one test over one window, and the page describes its limits; it does not say copying cannot work.

How do you test a strategy without risking money?

Some of the figures above say whether fees are in them: the Kalshi 70¢ result is after fees, the Kalshi 95¢ and 5¢ arithmetic is before fees, and our forward test includes the fee. Our own break-even arithmetic is before fees, and the other figures do not say, so read them as rough. You can get your own for one strategy. Whalidate books virtual copies of whale trades at real Polymarket prices with the taker fee taken out of each result, so you can look at the average and the worst stretch before any money is involved. It covers copying whales, not market making, arbitrage or news trading, and the page says what a paper result can and cannot tell you. A good paper result is not a promise, and a bad one costs nothing.

Whalidate is not affiliated with Polymarket. Nothing here is financial advice.

Questions people ask

What is the best Polymarket strategy?

The sources we read do not name one. Galaxy Research found 69.2% of retail accounts on polymarket.com below break-even (1 October 2026), and a Yale and London Business School study found about 3% of accounts trade with skill. Each strategy on this page has a cost, and whether any of them beats that cost depends on the market, the price and the period, so a test with fees included tells you more than a ranking.

Is arbitrage on Polymarket risk-free?

No. A set of outcomes that does not add up to $1 can leave a gap, but each leg taken from the book pays the taker fee, books can be thin so a later leg fills at a worse price, and a market that is disputed can hold your money for 4 to 6 days. Across venues, Gebele and Matthes found persistent gaps averaging 2 to 4% between equivalent markets and attribute them to structural frictions such as different event definitions, so two similar titles can be two different bets.

Can bots trade on Polymarket?

Yes. Polymarket's documentation describes public order-book and data endpoints, and orders are signed before they are submitted. Galaxy Research found about 4.1% of accounts automated (using its own cut-off of 50 orders per active day), placing 80.8% of orders on polymarket.com, and those accounts finished up $246.8 million in total. That is a report of what happened to them, not a forecast for a new bot.

Are 99¢ shares free money?

No. A share bought at 99¢ pays $1.00 if it wins, which is 1¢ of profit, and can lose the whole 99¢ if it does not. In one week of our own data (4 to 11 October 2026), shares bought at 98¢ or more ended in profit 87.4% of the time and returned -2.0%. A taker also pays the fee, and a disputed market can hold your money for days.

Sources