Polymarket Arbitrage Explained (2026): Complete Sets, Multi-Outcome Markets and Cross-Venue Gaps

"Arbitrage" is the most overused word in Polymarket marketing. Many GitHub repos and Telegram channels call almost any automated strategy an arbitrage bot. True arbitrage means locking in a profit no matter how the event turns out, and on Polymarket it comes in a few specific forms. Each one is narrower than it used to be, because Polymarket now charges taker fees on most categories and because hundreds of bots compete for the same gaps.

This guide covers the four real kinds of Polymarket arbitrage, the fee math that decides whether a gap is worth anything, and the risks that turn "risk-free" into "unexpectedly risky". It's part of our Polymarket trading bot guide.

The building block: $1 complete sets

Every binary Polymarket market has a YES token and a NO token. Exactly one of them pays $1 at resolution, and the other pays $0. Polymarket's contracts let you:

• Split 1 pUSD into one YES plus one NO.
• Merge one YES plus one NO back into 1 pUSD, at any time, without waiting for resolution.
• Redeem the winning token for $1 after resolution.

These operations are documented in How Positions Work. Because YES + NO is always worth exactly $1, the prices of the two sides are tied together. When the order books drift away from that tie, an arbitrage exists, at least on paper.

Type 1: Complement arbitrage (YES + NO ≠ $1)

Buy-side version. If the best ask for YES plus the best ask for NO adds up to less than $1, buy both and merge them back into $1 immediately. The profit is the gap, minus costs.

Sell-side version. If the best bid for YES plus the best bid for NO adds up to more than $1, split $1 into YES + NO and sell both.

The fee math that kills most of these

Polymarket's taker fee per share is feeRate × p × (1 − p). The rate is 0.04 for politics, finance, tech and mentions, 0.05 for sports, economics, culture, weather and other, 0.07 for crypto, and 0 for geopolitics (fees). If you take both legs, you pay the fee twice.

Example: a sports market where YES asks 47¢ and NO asks 51¢. The gross gap is 2¢.

Leg	Price	Taker fee per share (0.05 × p × (1−p))
YES	$0.47	$0.0125
NO	$0.51	$0.0125
Total cost per set	$0.98	+$0.0250 → $1.0050

That apparent 2¢ arbitrage loses half a cent per set. Near 50/50 prices, a complete set taken on both legs costs roughly feeRate × 0.5 in fees: about 2¢ in politics, 2.5¢ in sports and 3.5¢ in crypto. The gap has to exceed that before slippage. Only fee-free geopolitics markets and maker orders change the picture.

Making instead of taking. Makers pay no Polymarket fee and can earn maker rebates. You can try to build the set with resting orders on both sides. The catch is legging risk: one side fills, the other doesn't, and you're left holding a directional position. That's market making with extra steps, and it needs inventory management.

Volume rebates. Heavy takers can earn a share of their fees back through the Taker Rebate Program. Tiers range from 3% (Bronze) to 50% (Obsidian), based on 30-day weighted volume. This lowers the bar for very active bots, but not for a retail account.

Type 2: Multi-outcome (negative-risk) arbitrage

Events with several mutually exclusive outcomes ("Who will win the election?") are built as groups of binary markets. Polymarket links them as a negative-risk group: exactly one market resolves YES, and a NO share in one outcome can be converted into one YES share in every other outcome (Negative Risk Markets).

That creates two arbitrage shapes:

• Sum of YES asks < $1. Buy one YES in every outcome. Exactly one pays $1.
• Sum of YES prices > $1. The mirror image: buy NO in every outcome. With N outcomes, N − 1 of those NO shares pay $1, so if the NO asks add up to less than N − 1 dollars, the set is underpriced. (Conversion lets you turn NO shares into YES shares on the other outcomes if that's the cheaper route.)

Why these gaps exist: a group with ten or twenty candidates has many thin books, and the prices don't update together. Why they're hard to capture:

1. Fees on every leg. Twelve legs means twelve taker fees, although the fee on each low-priced longshot is small because p × (1 − p) is small near 0.
2. Depth. The cheapest ask on a 2¢ longshot may be good for only a few dollars.
3. Placeholders and "Other". Some events use augmented negative risk, with placeholder outcomes that get named later and an "Other" bucket whose definition changes. Polymarket explicitly warns to trade only named outcomes and to avoid trading "Other" directly (docs). A complete set built on a changing outcome list isn't complete.
4. Speed. Specialised bots scan every neg-risk group continuously.

Type 3: Cross-venue arbitrage (Polymarket vs Kalshi and others)

The same real-world question often trades on Polymarket and on Kalshi (or Polymarket US, Gemini, Crypto.com and others). If YES on one venue plus NO on the other costs less than $1 after both venues' fees, you appear to lock in the gap.

This is the version that looks most like free money and carries the most hidden risk:

• Resolution wording differs. Different sources, cutoff times, treatment of edge cases (postponements, recounts, revised data). If one venue resolves YES and the other resolves NO on the "same" question, both legs lose. Read both rule sets line by line. Mithril's cross-venue write-up and gobull.ai's guide make the same point.
• Two fee schedules. Kalshi's standard taker fee is round up(0.07 × C × P × (1 − P)) per order (Kalshi fee schedule). Polymarket's depends on the category.
• Capital in two places, locked until both resolve. That return has to beat simply holding cash.
• Access. Kalshi is a US-regulated exchange with its own eligibility rules. The international Polymarket platform is close-only for US users (geoblock). Very few people can legally hold both legs. Never use a VPN to get around a venue's restrictions.
• Settlement timing. One leg can pay days before the other, or a dispute on Polymarket can delay one side for several days (see how Polymarket markets resolve).

For a broader platform comparison, see Kalshi vs Polymarket.

Type 4: Logical (correlated-market) arbitrage

Some markets constrain each other logically:

• "Candidate X wins the presidency" can't be more likely than "Candidate X wins the nomination".
• "BTC above $150k on Dec 31" can't be more likely than "BTC above $120k on Dec 31".
• "Team wins the championship" can't be more likely than "Team reaches the final".

When the prices break these constraints, you can buy the cheap side of the constraint and sell the expensive side for a position that can't lose, or can only lose in odd edge cases. These trades are rarer and need careful reading of each market's rules: are the dates, sources and definitions really nested? They're also where human judgement still beats a naive scanner.

What "arbitrage bots" on 5-minute crypto markets actually do

Polymarket lists rolling "Up or Down" markets on BTC, ETH, SOL, XRP and others, in 5-minute, 15-minute, hourly, 4-hour and daily windows. Many GitHub projects advertise "arbitrage" on these markets. Most of them compare Polymarket's odds with a live exchange price and buy the side that looks underpriced. That's a directional latency trade, not an arbitrage. If the price reverses before the window closes, it loses.

The constraints are tight here: crypto markets carry the highest taker fee rate (0.07), selected up/down markets add a 250 ms taker delay, and the 5- and 15-minute markets resolve on a Chainlink TWAP stream, not on the exchange price a bot is watching. We explain all of it in Polymarket crypto up/down markets.

Risks in "risk-free" trades

Risk	What happens	Mitigation
Legging	One leg fills, the other moves away	Fill-or-kill orders on both legs, small size, pre-check depth
Fees miscounted	Profit turns into loss after two or more taker fees	Compute the fee per leg with the market's actual rate
Resolution mismatch	Cross-venue legs resolve differently	Compare rule texts; skip anything ambiguous
Disputes and clarifications	Funds locked for days; rare 50/50 resolutions	Size for the lock-up; avoid contentious markets
Tick size changes	Orders rejected when a price crosses 96¢ or 4¢	Handle tick_size_change events (websocket notes)
Engine restarts	Two minutes of post-only mode after maintenance	Back off, retry as post-only (docs)

Is Polymarket arbitrage still worth it in 2026?

For most retail traders, not as a main strategy. The clean versions (complete sets, neg-risk groups) are mostly arbitraged away within moments by well-capitalised bots, and taker fees have raised the break-even gap. The messy versions (cross-venue, logical) pay better but carry risks most "risk-free" guides leave out. CopyGrade puts it well: arbitrage on Polymarket is mostly "a reason prices stay coherent, not an open seat at a table" (CopyGrade).

Arbitrage is still useful to understand. It explains why prices behave as they do, and the same fee math applies to every other automated strategy, including copy trading and Kelly-sized directional trades.

FAQ

Is Polymarket arbitrage legal? Trading price differences is ordinary market activity. Using a platform from a jurisdiction where it's restricted isn't allowed. See Is Polymarket legal?

Do I have to wait for resolution to collect a complete-set arbitrage? No. On Polymarket, a matched YES + NO pair can be merged back into pUSD at any time.

What's the minimum gap worth trading? More than the sum of taker fees on every leg you take (about 2–3.5¢ per set near 50/50 prices, depending on category), plus slippage and a margin for failed legs.

Does Ghost Trader do arbitrage? No. Ghost Trader runs copy trading and a crypto up/down strategy, with fee-aware edge thresholds and hard risk limits. We don't sell it as risk-free, because it isn't.

---

Ghost Trader automates Polymarket trading with your own wallet on your own machine. It only acts when a signal clears your minimum edge after costs, sizes positions with fractional Kelly, and stops at your daily-loss limit. $199/month with paper mode included. See pricing.

Not financial advice. Trading can lose money. Ghost Trader isn't available where Polymarket restricts trading, including the US, and is not affiliated with Polymarket.