The world’s first prediction market for private, unlisted companies.

NOTSTOCKS
Sign in
Back to blog

1.4¢ Gaps Shrink to 0.6¢: Prediction Market Arbitrage for Traders 2026

Prediction market arbitrage title card

Arbitrage opportunities do exist in prediction markets, but they are usually short-lived and rarely truly risk-free after execution costs. Capturing them requires automation, low-latency execution, and careful risk controls. Once bid-ask spreads and transaction costs are counted, both academic and practitioner evidence show the edge shrinks fast, and much of what looks like free money is actually a race against bots.

TL;DR:

  • Arbitrage opportunities are usually short-lived, heavily impacted by execution costs like spreads, fees, and transfer expenses, which often eliminate the profit margin.
  • Market-rebalancing arbitrage occurs within a single market, while combinatorial arbitrage exploits dependencies across multiple markets; matching settlement terms is crucial for both.
  • Trading on prediction markets requires careful scanner designs that account for order-book depth, semantic differences, and platform-specific mechanics to avoid false positives and execution failures.
  • Speed and automation are key to success, as most profitable arbitrage falls to bots that act faster than manual traders, capturing rapid, fleeting gaps before they close.
  • Liquidity at the quoted price, not just the displayed gap, determines whether an arbitrage opportunity is tradable, especially on lower-volume or private markets.
Notstocks
Explore Private Market Signals
Explore prediction markets across private companies and emerging sectors, with access designed for more informed trading decisions.
Explore NotStocks

Table of Contents

Market-Rebalancing vs Combinatorial Arbitrage

Prediction market arbitrage falls into two categories, and confusing them causes most scanner failures.

Market-rebalancing arbitrage happens within a single market when the sum of prices across all mutually exclusive outcomes drifts away from the value that guarantees a payout. If a three-outcome market prices its shares at $0.30, $0.35, and $0.30, buying one share of each costs $0.95 for a bundle that always pays $1.00. Combinatorial arbitrage works across separate but logically dependent markets: a market asking “will Team A win the tournament” and another asking “will Team A win their group” are linked, since winning the tournament requires winning the group first, so a portfolio can be built that guarantees at least one leg pays off regardless of outcome.

  • Market-rebalancing arbitrage exploits a pricing gap inside one market’s own outcome set.
  • Combinatorial arbitrage exploits a logical dependency between two or more separate markets.
  • Both require settlement terms to match exactly, since a subtle difference in resolution wording can turn a guaranteed payout into an open bet.

The Dagstuhl/AFT (2025) framework formalizes both types and is a useful reference for anyone building a scanner rather than eyeballing prices.

Why Execution Costs Erase Most Apparent Edges

A mispricing on a screen is not a profit until it survives the full cost stack. Spreads, taker fees, funding costs on capital held until settlement, and gas or transfer fees on-chain all cut into the gross gap before a trade is even confirmed.

  • Bid-ask spreads alone can consume most of a small pricing gap on lower-volume markets.
  • Funding and transfer costs apply when capital sits locked across two platforms until resolution.
  • Partial fills, latency, and order-size limits mean the “priced” opportunity is often smaller than the one you can actually execute.

Cao (2013) found average returns on arbitrage bundles fall from roughly 1.4 cents to about 0.6 cents once bid-ask spreads and transaction costs are factored in, and the probability of a profitable trade drops substantially alongside it. Practitioner analysis from PredictReport echoes this: cross-platform arbitrage only pays after commissions, slippage, funding, transfer, gas, and settlement costs are all subtracted, and the gap tends to close before a manual trader can act on it.

Detecting Opportunities With Scanners and Data Models

Finding a genuine arbitrage window before it closes is mostly a data problem, not a trading problem.

  1. Bundle-sum checks flag any market where outcome prices across a single event fail to sum to the payout value.
  2. Cross-platform price-gap scanners compare equivalent contracts on different venues to catch mispricings between ecosystems.
  3. Dependency-graph scanners map logically linked markets (nested events, conditional outcomes) to surface combinatorial opportunities that a simple bundle check would miss.

Each needs live order-book depth, fill history, and settlement-rule metadata pulled through the platform’s API, since price alone hides whether the size you need is actually available. The biggest source of false positives is semantic: two contracts that read as equivalent can resolve on different triggers or timestamps. The Flashbots collective notes that scanners need a parsing layer that converts human-readable resolution text into standardized predicates, or they will flag pairs that are not truly equivalent.

Pro Tip: Build your dependency scanner to diff resolution-source text and deadline fields programmatically, not just outcome labels, since two “yes/no” markets with identical names can still settle on different events.

Execution Tactics That Protect the Hedge

An arbitrage trade is only hedged once both legs are filled. Until then, it is a directional bet with extra steps.

  • Mix limit and taker orders: post limits for the illiquid leg and take the liquid one to reduce the odds of a one-sided fill.
  • Size positions to the thinner leg’s available depth rather than the headline price, and slice large orders instead of sending them at once.
  • Set automatic unwind thresholds that close the filled leg if the second leg does not execute within a defined window.
  • Keep a reconciliation log matching timestamps on both legs, since non-atomic order books can leave you holding only one side.

Before running any of this live, dry-run the bot against historical order books, monitor latency between your feed and the venue, and keep a settlement-timing calendar so capital is not locked past an expected resolution date.

Pro Tip: Treat every arbitrage bot as a directional-risk system first and an arbitrage system second, because the failure mode that matters is the one where only one leg fills.

What the Research Says About Realized Profits

The published evidence points the same direction: arbitrage exists, but it is thin and shrinking.

  • Cao (2013) confirmed the existence of arbitrage in OCR prediction markets, but showed magnitude and probability of profit both fall sharply once real spreads and costs are applied.
  • The Dagstuhl/AFT (2025) paper counted many short-lived opportunities but stressed that non-atomic order books introduce execution risk: a trader can buy one leg and fail to sell the other before the price moves.
  • GetArbitrageBets reports that realized profits exist historically but are not a persistent, risk-free yield, since spreads and fees compress returns and a failed execution can convert a hedge into an open position.

Reporting from Yahoo Finance describes bots and automated systems capturing large profits on major platforms, which underscores why manual retail arbitrage has become a harder game than the academic papers alone suggest.

Arbitrage in Lower-Liquidity Private-Company Markets

Private-company prediction markets typically carry thinner order books and more bespoke settlement terms than sports or election markets, which raises the odds of a partial fill turning a planned hedge into a naked position. NotStocks publishes live price history and daily research notes on each listed company, which helps with the semantic matching step before a trade rather than after a bad fill. In practice, that means treating pre-trade research as mandatory and sizing trades smaller than the headline liquidity suggests.

Surveillance, Insider Information, and Platform Rules

Regulatory attention on event contracts has intensified, and it directly affects how arbitrage strategies should be designed. A 2026 industry summary from Willkie describes proposed rule changes that emphasize contract drafting, settlement-term clarity, and market-integrity surveillance, including a structured review process with a defined timeline for challenges to a listed contract.

For arbitrageurs, three things follow from this. First, settlement semantics matter as much for compliance as for trade safety: a contract with ambiguous resolution language is more likely to trigger regulatory review, which can freeze or delay payout and turn a hedged position into locked capital. Second, surveillance systems on major platforms are built to catch coordinated or wash-style trading patterns, so automated arbitrage bots need to operate within each platform’s stated rules on order behavior rather than assuming arbitrage is exempt from scrutiny. Third, insider information remains a distinct risk from arbitrage itself: trading on non-public knowledge of an outcome is a different activity from exploiting a public pricing gap, and platform terms typically treat the two very differently even though both can look like “beating the market” from the outside.

None of this means arbitrage is prohibited. It means the operational checklist for a live strategy has to include reading each platform’s rulebook on manipulation and insider trading alongside the purely financial cost stack, because a technically profitable trade that violates platform rules can be reversed or penalized regardless of the pricing logic behind it.

Lessons From Past Arbitrage Trades

The clearest successes in prediction market arbitrage have come from automated systems working at a scale manual traders cannot match. Bot operators who built cross-platform scanners early captured large, repeated profits before spreads compressed, a pattern documented in reporting on bots dominating major platforms with sizable gains. The common thread in successful trades is speed: the gap was priced correctly by the scanner, sized to available depth, and closed within a single fill cycle.

Failed trades follow a different pattern almost every time: a trader identifies a bundle-sum or cross-platform gap, executes one leg, and then finds the second leg has moved or dried up before the order clears. The Dagstuhl/AFT (2025) paper frames this as a consequence of non-atomic order books: there is no mechanism guaranteeing both legs of an arbitrage execute together, so the trade can silently become directional. Another recurring failure mode is semantic mismatch, where two markets that appeared to be equivalent settle on different triggers or dates, and the “guaranteed” payout never materializes because the positions were never truly offsetting.

Illustration of failed two-leg arbitrage

The practical lesson is not that arbitrage cannot work. It is that every historical failure traces back to either an execution gap between the two legs or a settlement-rule mismatch that was missed during scanning, both of which are the exact problems that automated dependency-graph scanners and unwind rules are built to catch.

How Liquidity and Volume Shape the Opportunity

Liquidity is the variable that decides whether a pricing gap is theoretical or tradable. A market with deep order books at the quoted price lets a trader execute both legs of an arbitrage near the displayed spread, while a thin market can show an attractive gap on paper that vanishes the moment a real order tries to fill it, because the available size at that price is a fraction of what the screen implies.

Higher-volume markets also tend to close arbitrage gaps faster, since more participants and bots are watching the same pricing relationships, which is part of why PredictReport frames cross-platform arbitrage as a race that closes as soon as enough traders act on it. Lower-volume or niche markets, including many private-company and long-tail event markets, can hold a mispricing longer simply because fewer systems are scanning them, but that same thinness means position sizing has to be conservative and partial fills are more likely.

The practical rule is straightforward: liquidity at the quoted price, not the headline price gap itself, determines whether an arbitrage opportunity is real. A scanner that ranks opportunities by gap size alone, without weighting by order-book depth, will consistently surface trades that cannot be executed at the size needed to make them worthwhile after costs.

Arbitrage Strategy Differences Across Platforms

Arbitrage design has to adapt to each platform’s mechanics, and treating all prediction markets as interchangeable is a common mistake. Some platforms settle on-chain, which adds gas costs and transfer delays that a purely price-based scanner will underweight, while others settle off-chain with faster but less transparent processes. Fee structures also vary: a platform with a flat trading fee changes the breakeven math differently than one with a variable spread-based cost, so the same nominal price gap can be profitable on one venue and unprofitable on another.

Order-book structure differs too. Some platforms use continuous limit order books similar to traditional exchanges, which support the taker/limit mixing tactics described earlier, while others use automated market-maker style pricing where the act of trading itself moves the price, meaning slippage has to be modeled explicitly rather than assumed away. Cross-platform arbitrage strategies, like those PredictReport describes between major venues, only work when a trader accounts for these mechanical differences alongside the raw price gap.

Comparison of prediction market platform mechanics

For niche or specialized ecosystems, such as markets built around private, unlisted companies, settlement terms and listing structures can be more bespoke than on large general-purpose platforms, which raises the value of platform-specific research before assuming a strategy that works on one venue transfers cleanly to another.

Where Arbitrage Fits in a Trader’s Toolbox

Arbitrage suits systematic, automation-first operations more than discretionary trading, and researchers often use it as a lens for testing market efficiency rather than as a standalone income strategy. Many setups that look like arbitrage are really directional bets wearing a hedge.

— Max

Researching Private-Company Markets With NotStocks

The same detection discipline that applies to sports and election markets applies to private-company markets, where thinner books make pre-trade research even more valuable. Some platforms focus exclusively on private, unlisted companies and pair live price and trade history with daily research-style notes, enabling users to check a listing’s settlement basis before sizing a position rather than after a partial fill.

Notstocks

  • Browse live price history and daily notes on listed companies through the Learn hub.
  • Submit or claim a company listing to track a market you want to monitor for pricing gaps.

If you want to test how settlement terms and pricing behave on a lower-liquidity, private-company market, start on the NotStocks platform page and review a listing’s history before placing a trade.

Sources

FAQ

Is it possible to arbitrage prediction markets?

Yes, mispricings between outcomes or across platforms do occur, but they tend to close quickly once spreads, fees, and settlement risk are factored in. Cao (2013) confirmed the existence of these opportunities while showing their average value shrinks sharply after real transaction costs are applied.

Exploiting a public pricing gap between two legitimate markets is a different activity from trading on non-public information, and platforms generally treat the two very differently in their rules. Regulatory guidance summarized by Willkie (2026) points to settlement clarity and surveillance as the areas platforms and traders should watch, rather than treating arbitrage itself as prohibited.

Can you really make money with arbitrage?

Realized profits exist, but practitioner analysis from GetArbitrageBets shows they are compressed by fees and spreads and often captured by automated systems before a manual trader can act. Reporting on bots dominating platforms with large profits, as covered by Yahoo Finance, underscores how much of the easy edge has already shifted to automation.

Is crypto arbitrage still profitable?

On-chain and crypto-adjacent prediction markets face the same cost pressures as any other venue, plus added gas and transfer costs, so profitability depends heavily on execution speed and order-book depth at the moment of the trade. Practitioner guidance from PredictReport treats these costs as standard inputs to the breakeven calculation rather than exceptions to it.

Written with BabyLoveGrowth for Google and AI search