Business

Article series

Prediction Markets

Prediction Markets Are Growing. What Do the Numbers Actually Prove?

By iGaming Review · Published 30 October 2026

Trading volume can show activity without proving new demand, reliable liquidity or a sustainable business. This analysis sets out the evidence operators should request before committing their customers, capital or brand.

iGaming Review · Analysis · Research current to 21 October 2026

A prediction-market platform can report more trading without attracting proportionately more customers. A busy contract can still be expensive to enter or exit. And a probability quoted to one decimal place can conceal a market with very little money available at that price.

For operators assessing a partnership, investors comparing platforms and publishers deciding which odds to quote, those distinctions matter. The useful question is what evidence connects activity to a durable business and an informative market.

October puts the numbers under scrutiny

The Wall Street Journal reported on 17 October that about $108 million of ADI Predictstreet’s $167 million in trading through late September came from wallets apparently linked to the company. Gibraltar’s gambling commissioner told the newspaper that a preliminary review found compliance with local self-trading rules and no consumer detriment. ADI Predictstreet did not respond to the Journal’s requests for comment; FIFA declined to comment. These are reported findings and responses, not an established breach.¹

Our News report on the FIFA partner provides the immediate context. The wider commercial question is how a prospective partner distinguishes independent customer demand from other sources of activity. Neither a headline total nor a wallet count settles that question.

Another October development points towards better disclosure. Kalshi announced low-volume labels, historical forecast scoring and more prominent trade feeds for its midterm election markets on 6 October, as covered in our report on its transparency features. The changes provide more context; whether they improve decisions is a separate question.²

Start with the unit being counted

Before comparing two growth rates, ask for their definitions. Is the figure a count of contracts traded, their face value, cash exchanged at transaction prices, or fees collected? Does it cover the same period and product types? Does the source count a matched trade once, or aggregate both sides?

The separation is visible in the underlying infrastructure. Kalshi’s market data exposes volume and open interest as distinct fields. CME’s explanation of derivatives statistics likewise distinguishes trading during a period from contracts that remain outstanding. Open interest is a stock of positions; volume measures a flow of transactions.³⁴

Repeated resale can be entirely legitimate. It enables traders to adjust positions and others to take the opposite view. The analytical mistake is to treat each resale as another customer arriving with new money.

A useful comparison therefore reports turnover alongside outstanding positions, active customers and net customer funding, with the counting method attached. Combining event contracts with other products, or comparing an exchange’s face-value turnover directly with sportsbook stakes, can obscure the business being measured.

Count independent participation, then test its durability

Accounts are an operational measure. Independent customers are a commercial one. A platform assessing its own demand should be able to explain how it handles multiple accounts, professional liquidity providers, affiliated entities and shared funding arrangements. An external observer may lack the information needed to identify the ultimate controller of a wallet.

That uncertainty cuts both ways. Connected funding deserves examination, but does not by itself establish unlawful trading. Nor should legitimate market making be discarded as worthless activity: participants posting buy and sell orders can help other customers transact.⁵

Our proposed commercial test is to follow customer groups over time. How many first-time funded customers return after 30 or 90 days? What share of activity comes from the largest participants? What remains after a major sporting event ends or an acquisition incentive expires?

These are questions for a data room, not figures that can be reconstructed confidently from a promotional chart. Where the evidence is unavailable, the honest conclusion is that the claim cannot yet be assessed.

A quoted probability is not an executable price

Liquidity concerns what a customer can trade now, at what size and cost. Kalshi’s order book displays resting orders with their prices and quantities. A narrow gap between the best buy and sell prices is useful, but a partner should also ask how much can actually be executed near those prices.⁶

Polymarket’s documentation provides a concrete distinction. Its displayed price is normally the midpoint between the best bid and ask; when that spread exceeds $0.10, it displays the last traded price instead. Neither convention guarantees that the displayed number is available for an immediate transaction.⁷

The 68.6¢ figure is the average execution price, not a new displayed probability. Buying YES consumes progressively more expensive sell orders; selling YES into a thin book can push execution prices down. A limit order can cap the price paid, but may leave part or all of the intended purchase unfilled.

For an operator considering distribution, a practical test would examine execution at several predetermined order sizes, during ordinary trading and around major news. For a publisher quoting a probability, the equivalent discipline is to record the time, market definition and liquidity context. A clean percentage should not imply precision the available evidence cannot support.

Liquidity and the information behind the price

The “wisdom of crowds” argument concerns the aggregation of dispersed information. Research by Justin Wolfers and Eric Zitzewitz examines how markets can turn such information into forecasts. A trade count does not tell us how many independent judgements contributed to a price.¹⁰

If most activity comes from platform-controlled algorithms, headline volume cannot substantiate a broad, independent crowd: many transactions may recycle the same assumptions. That does not automatically make the price wrong. Models and professional market makers can incorporate useful external information. The relevant questions are what information drives the quotes and whether informed outsiders can challenge them at realistic cost.

Thin depth can make that challenge expensive. A price move caused by one order consuming scarce liquidity need not reflect new information about the underlying event. Conversely, a deep book supplied by one firm does not itself establish a diversity of views or accurate forecasting.

Our assessment would therefore separate trading conditions from evidence of predictive performance. A claim to outperform polling needs testing against a relevant polling-based forecast for the same outcome and forecast horizon. Neither turnover nor the number of automated trades can settle that comparison.

Forecast quality and commercial value need separate tests

Kalshi says the scores displayed on its midterm markets draw on historical data, weighted by volume and time to resolution. They cannot establish the accuracy of a forecast for an election that has not yet occurred. A Brier score assesses probability forecast error; calibration asks a related but different question, such as whether outcomes assigned around 70% probability occur around 70% of the time across an appropriate sample.²⁸

A credible evaluation should disclose which markets were included, when forecasts were recorded and the benchmark used. A good last-minute score should not be presented as evidence of equally good forecasting months earlier. Low volume does not automatically make a forecast wrong; high volume does not make it certain.

The commercial assessment is different again. Polymarket documents fees and programmes that return a portion to liquidity providers and eligible takers. Gross fees therefore need to be reconciled with rebates and incentives before a reader can judge the economics.⁹

For a house-banked sportsbook or casino, gross gaming revenue (GGR) broadly means stakes less player winnings, before operating costs. Kalshi instead identifies transaction fees as its revenue model.¹¹ Liquidity providers may earn the spread; it is not automatically platform revenue.⁵

For an incumbent operator, moving customers to a peer-to-peer exchange is therefore more than a cross-sell. Familiarity with fixed odds does not establish demand for tradable contracts, explicit fees and execution that depends on available counterparties. Our commercial test would ask whether the operator’s share of recurring fee revenue, net of incentives, covers conversion and support costs—and whether it adds value or displaces existing gaming revenue.

Our proposed partner assessment would request a bridge from gross fees to amounts retained, identify revenue-sharing obligations, and then consider acquisition, support and compliance costs. Those disclosures would support a business judgement. Trading volume alone cannot supply them.

Market access depends on jurisdiction

The commercial assessment also carries jurisdiction risk. Kalshi operates as a CFTC-designated derivatives exchange.¹² Polymarket requires a distinction: its US exchange, QCX LLC, is a CFTC-regulated designated contract market, while its separate international platform is not CFTC-regulated.¹³ ADI Predictstreet’s operator, Predict Street Limited, appears on Gibraltar’s register of authorised prediction-market operators.¹⁴

Those statuses do not establish permission to serve customers everywhere. In Great Britain, the Gambling Commission says current prediction-market models would appear to fall within betting-intermediary licensing, depending on their design; operators without the appropriate licence should not target or transact with consumers there.¹⁵ Our assessment would therefore identify the legal entity, product and customer location, and establish the applicable licensing and consumer-protection requirements before treating an existing player database as an addressable market.

Earlier in this series, we examined how institutional money and political pressure reinforce each other. The next commercial test is whether the evidence behind a platform’s growth can withstand the same scrutiny as its legal position. A prospective partner should ask for that evidence before giving the platform access to its customers or lending it the credibility of its brand.