← Research
Operator Research Prediction Markets Pricing 16 min read • March 2026

Who Sets Prediction Market Odds — and What Operators Can Learn

Kalshi hit $50 billion in annualized volume in 2025. Wall Street market makers are now the de facto oddsmakers in the fastest-growing sports wagering category. Here is what traditional operators must build to stop being price-takers.

By the Metrics
165x
Kalshi volume growth, $300M → $50B annualized (2024→2025)
69.86%
Calibration vs. accuracy model performance gap (NBA study)
~55%
Institutional share of prediction market funding
Problem
Prediction market odds are increasingly set by Wall Street market makers — not sportsbook operators — leaving traditional operators as price-takers in a category they helped create.
Approach
We analyze the microstructure of prediction market pricing: who the real liquidity providers are, how calibration beats accuracy, and where operator data gives a structural edge over financial firms.
📈
Outcome
Operators who build proprietary calibration layers, acquire exchange infrastructure, and deploy sub-second re-pricing can reclaim margin — and use prediction markets as a customer acquisition vehicle in states without regulated sports betting.
in 𝕏

The question of who actually sets odds in prediction markets is more consequential than most operators realize. Traditional sportsbooks built an entire industry on the assumption that the house sets lines and manages risk. In prediction markets, that model is inverted: prices emerge from peer-to-peer trading, and the entities doing the most to anchor those prices are not operators at all. They are institutional market-making firms from Wall Street.

This structural shift has direct implications for margin. Kalshi and Polymarket now command 97.5% of the US prediction market, and the speed at which this category is scaling means the window for operators to claim a price-setter position is closing. Understanding the microstructure — who provides liquidity, how prices form, and where proprietary data creates an edge — is now a strategic requirement, not an academic exercise.

No House Edge: How Prediction Market Prices Actually Form

In a traditional sportsbook, a trader or algorithmic model sets the opening line. Risk is managed by adjusting spreads or restricting sharp bettors. The house maintains a margin on every bet. Prediction markets work on an entirely different mechanism.

Prediction market prices are set through a central limit order book (CLOB) — the same mechanism used in equity markets. Buyers and sellers submit bids and offers at prices they believe reflect the true probability of an event. Contracts trade between $0.01 and $0.99, where the price represents the implied probability. A contract trading at $0.62 means the market collectively believes there is a 62% chance the event occurs. Winning contracts settle at $1.00.

There is no house setting a line. Price formation is emergent — it is the aggregate result of every market participant's view on probability. This creates two important dynamics: first, price discovery is faster and more responsive than any single trader's model; second, whoever provides the most liquidity has the greatest influence over where prices anchor.

That liquidity anchoring role is now dominated by institutional players, not operators. When retail flow is thin — which is most of the time in newer or lower-volume event categories — firms like Susquehanna International Group, Jump Trading, and Galaxy Digital fill the bid-ask spread and effectively set the market price.

US Market Share
97.5%
Kalshi + Polymarket combined, as of 2025 (KuCoin)
Weekly Volume Record
$5.35B
Combined notional volume, March 2–8 2026 (DeFiRate)
Kalshi Annualized Volume
$50B
Up from ~$300M the prior year — approximately 165x growth in a single year

The probability framing of contracts — expressed as a price between $0.01 and $0.99 rather than fractional or American odds — is a meaningful UX difference. For bettors who think in percentage terms, the cognitive barrier is lower. A $0.65 contract on a team winning feels more intuitive than −185 on a moneyline. This accessibility is one of several reasons volume has scaled so rapidly.

Wall Street Moved In: The Institutional Takeover of Sports Pricing

The composition of capital in prediction markets is not what most sportsbook executives assume. Approximately 55% of prediction market funding comes from institutional sources, with Susquehanna International Group and Jump Trading serving as primary market makers on Kalshi. These are not casual traders with sports opinions — they are algorithmic desks running the same co-location and latency optimization strategies they deploy in equity and options markets.

What this means in practice: the financial market microstructure has been imported directly into sports event pricing. Cross-platform arbitrage opportunities — pricing discrepancies between Kalshi, Polymarket, and regulated sportsbooks — close within seconds and typically yield 0.5–3% returns per trade. Capturing those returns requires sub-10 millisecond infrastructure. Any operator entering the prediction market space without that capability is systematically disadvantaged from the moment a line moves.

The underappreciated battle: Industry insiders describe the oddsmaking competition on Kalshi and Polymarket as fundamentally underappreciated. Who anchors prices in sports prediction markets will determine whether traditional operators or Wall Street firms extract the majority of margin from this emerging category. Operators entering as mere front-ends — without building pricing capability — are conceding that margin permanently.

The scale of what is at stake is visible in Kalshi's financials. The platform generated $263.5 million in fee revenue in 2025, the majority from sports event contracts. Its December 2025 Series E valued the company at $11 billion — and both Kalshi and Polymarket are reportedly targeting $20 billion valuations in 2026. Monthly active users on Kalshi grew 8.5x in a single year, from 600,000 to 5.1 million.

This growth is not incidental. It reflects a structural shift in how a segment of the betting public wants to interact with sports event probability. Operators who treat prediction markets as a niche or an experiment are misreading the trajectory.

Calibration Is the Right Target — Accuracy Will Lose You Money

The single most consequential technical insight for operators entering prediction markets is the distinction between predictive accuracy and calibration — and why optimizing for the wrong one systematically destroys margin.

Accuracy measures how often your model picks the correct outcome. Calibration measures whether your stated probabilities actually reflect true outcome frequencies. A model that says 60% and wins 60% of the time is well-calibrated, even if a different model picks winners more often. The difference sounds subtle. The financial impact is not.

69.86% Performance gap between calibration-optimized and accuracy-optimized pricing models in NBA studies — operators running accuracy-first stacks are systematically underpricing risk and gifting margin to sharp bettors and institutional arb desks

An NBA study comparing ML pricing approaches found that calibration-optimized models returned +34.69% while accuracy-optimized models returned −35.17% — a gap of 69.86 percentage points. The accuracy-first model picked winners more often but mispriced the margin on those wins and losses, resulting in systematic losses at scale. This is not a marginal difference. It is the difference between a profitable pricing operation and one that bleeds money to sharps.

The broader AI-driven modeling landscape reinforces why getting this right matters. Top AI models now beat closing line value (CLV) by 3–7% on average. Generative AI approaches reach 75–85% game-winner accuracy versus 50–60% for traditional statistical models. Operators who have not audited whether their existing pricing models are calibration-first or accuracy-first are almost certainly running the losing configuration without knowing it.

Model Optimization Target Simulated Return (NBA study) Risk Profile
Accuracy-optimized ML −35.17% Underprices tail risk; vulnerable to sharp exploitation
Calibration-optimized ML +34.69% Correct probability expression; sustainable margin
Top AI models (CLV edge) +3–7% vs. closing line Market-beating; requires real-time data integration

Sub-Second Re-Pricing Is Now Table Stakes

Even a perfectly calibrated model is only as good as its ability to update. In prediction markets, the speed at which prices respond to new information is a primary competitive variable — and the gap between institutional-grade infrastructure and operator-grade infrastructure is significant.

Breaking news can shift prediction market prices 40–50 percentage points instantly. A starting lineup change, an injury confirmed on social media, a referee appointment — any of these can render an existing price deeply mispriced within milliseconds. Sharp bettors and algorithmic funds have automated pipelines scanning these signals continuously. Operators whose re-pricing runs on scheduled batch updates are not just slow; they are systematically offering mispriced lines to the most sophisticated market participants.

The thin liquidity characteristic of newer event categories compounds this vulnerability. In a book with limited depth, a single $1,000 order can move a prediction market price 10 points. When that order comes from an informed bettor who just saw a news signal before the operator's system did, the operator is not just filling a bet — it is transferring margin directly to the sharpest player in the book.

The Precog signal: Precog, launched in Q1 2026 beta, assigns every prediction market wallet an “alpha score” based on historical hit rate, timing precision, and conviction sizing — the prediction market equivalent of sharp bettor profiling. The emergence of this tooling confirms that the market is sophisticated enough to systematically identify informed traders. Operators need analogous tools on their own books to identify and appropriately manage exposure from high-alpha accounts.

The implication is clear: event-driven re-pricing triggers, not batch updates, are the minimum viable infrastructure for any operator seeking to be a price-setter rather than a price-taker. This requires integrating news feeds, social media signals, and market data into an automated pricing loop that can update odds within milliseconds of a relevant event occurring.

Where Sportsbooks Beat Wall Street: Proprietary Data Moats

The picture so far is not encouraging for traditional operators. Institutional market makers have superior latency infrastructure, deeper capital reserves, and decades of experience running CLOB-based markets. But operators hold one structural advantage that financial firms genuinely cannot replicate: proprietary real-time sports data.

Susquehanna and Jump Trading are exceptional at financial microstructure. They are not exceptional at knowing that a starting midfielder is carrying a knock that did not make the official injury report, or that sharp money on a particular game has shifted the internal line by three points in the past two hours. Traditional operators accumulate this information continuously. Most are not yet deploying it systematically in prediction market pricing.

Flutter/FanDuel's CEO articulated the advantage directly: “The ability to price complex correlated outcomes accurately is something that we do every day in our core business.” Correlated outcome pricing — the ability to price a same-game parlay or a player prop conditional on the game result — is where years of sportsbook data create a genuine edge over generalist financial market makers.

Major operators are acting on this logic. Underdog acquired Aristotle Exchange, a CFTC-regulated exchange, to control the exchange infrastructure layer outright. DraftKings launched Predictions via CME Group and has announced plans to integrate Railbird. DraftKings CEO Jason Robins has explicitly identified prediction markets as the growth lever in the approximately 12 US states without regulated sports betting — a position now available in 38 states. FanDuel has similarly expanded nationwide, including California.

165x Kalshi’s volume growth from ~$300M to $50B annualized in one year — the window for operators to claim market-maker position in this category is narrowing fast, and front-end-only plays will not be sufficient

The strategic logic of exchange acquisition is straightforward: operators who own only the front-end are permanent price-takers. Operators who own the exchange layer can set fees, control market rules, and deploy their proprietary pricing advantage at the infrastructure level. Owning the stack is becoming a competitive necessity, not a growth option.

Pricing Gets You In. Personalization Keeps You Winning.

Prediction market pricing capability determines margin at the transaction level. But margin sustainability over time is determined by what happens after the first bet. This is where the operator's CRM and personalization infrastructure becomes the second half of the competitive equation.

The parallels between calibration in pricing and personalization in CRM are instructive. Just as a calibration-first pricing model expresses accurate probabilities and sustains margin, an AI-personalization-first CRM strategy expresses accurate relevance to each bettor and sustains engagement. Generic approaches in both domains systematically underperform.

The data on AI-driven personalization in sports betting is consistent: AI-personalized offers generate 20–30% higher revenue versus generic campaigns, and drive a +21% increase in bet frequency. For an operator using prediction markets as a customer acquisition vehicle in states without regulated sports betting, this personalization advantage compounds directly into lifetime value once a user is converted.

The prediction market format itself creates new personalization opportunities. Probability-framed contracts naturally lend themselves to explanatory content — operators can present “here is why this contract is priced where it is” narratives that are more engaging than a simple odds display. This is a CRM touchpoint, not just a pricing display. The lower cognitive barrier of percentage-based pricing means higher-frequency engagement from users who would disengage from a traditional odds presentation.

The broader market is splitting into two platform types: content-driven platforms that compete on novelty and engagement, and market-driven platforms that compete on liquidity depth, fee structures, and re-pricing speed. Operators need to compete on both dimensions simultaneously. The global AI sports betting market is growing from $9 billion in 2024 to a projected $28 billion by 2030 at a 21.1% CAGR — and the operators extracting the most value from that growth will be those who have combined pricing precision with personalization depth.

The Operator Checklist: From Price-Taker to Price-Setter

The steps below are not aspirational. They are the minimum viable actions for any operator that intends to be a pricing participant, not a pricing observer, in prediction markets.

Step 1: Audit your pricing stack for calibration vs. accuracy

Before deploying any capital in prediction markets, determine whether your existing models are optimized for calibration or accuracy. The 69.86% performance gap identified in NBA studies is not unique to basketball — it reflects a systematic problem with how most sports pricing models are evaluated. Replace accuracy-first models with calibration-first alternatives. This step costs nothing except analyst time and protects margin on every subsequent bet.

Step 2: Build sub-second re-pricing infrastructure

Scheduled batch re-pricing is not viable in a CLOB market. Move to event-driven triggers: news feed integration, social media signal processing, internal sharp money flow detection. The 40–50 percentage point price shifts that follow breaking news need to be processed by your system before they are processed by institutional arb desks. This is an infrastructure investment, not an optional upgrade.

Step 3: Layer proprietary data on top of any vendor-supplied odds

Vendor-supplied odds are by definition available to every operator using the same vendor. Proprietary data — injury intelligence, sharp money flows, player tracking feeds — is what creates pricing differentiation from institutional market makers who lack sports-specific data access. The operator who surfaces this data fastest and most accurately into their pricing model holds a structural edge that financial firms cannot replicate.

Step 4: Evaluate exchange infrastructure acquisition or partnership

Front-end access to prediction markets without exchange-layer control is a commodity position. CFTC-regulated exchange acquisition (as Underdog demonstrated with Aristotle) or CME Group partnership (as DraftKings demonstrated) provides the infrastructure control needed to extract maximum margin. Operators who defer this decision are not avoiding a cost — they are accepting permanent price-taker status.

Step 5: Connect prediction market engagement to CRM flows

Use AI personalization to drive repeat handle and cross-sell into regulated sports betting markets. The 20–30% revenue uplift from AI-personalized offers applies as directly to prediction market users as to traditional sports bettors. Build the CRM infrastructure to treat prediction market users as a distinct segment with distinct behavioral signals — probability-based engagement patterns, contract category preferences, and session timing differ materially from traditional bettor profiles.

The compounding advantage: Operators who execute all five steps create a compounding moat. Calibration-first pricing reduces adverse selection. Sub-second re-pricing closes the window for institutional arbitrage. Proprietary data creates pricing differentiation. Exchange ownership extracts infrastructure margin. And personalization converts volume into loyalty. Each layer reinforces the others. Operators who execute only one or two steps remain structurally vulnerable to the firms executing all five.

The category is scaling faster than any single operator's internal roadmap. Kalshi's 8.5x monthly active user growth from 600,000 to 5.1 million users happened in a single year. Combined weekly notional volume hit $5.35 billion in March 2026. The institutional capital is already in position. The question for operators is not whether to engage with prediction market pricing — it is whether they engage as price-setters or price-takers.

Data Sources & Attribution

Ready to Build Pricing Infrastructure That Competes?

BidCanvas provides calibration-first pricing models, sub-second re-pricing triggers, and sharp-money profiling tools built for operators entering prediction markets as price-setters — not price-takers.

Request Demo See the Product