Key takeaways
- The three largest venues recorded $45.33B in August 2026 volume, but that figure is a definition rather than a measurement, and the definitions in use do not agree with each other.
- Robinhood's own Q2 2026 release reported 13.6 billion event contracts through its hub and more than 3.5 billion on Rothera, then warned that the two cannot be added because each counts something different.
- The Block revised June 2026 upward by roughly $2B and July by roughly $2.4B after publication, while nobody was disputing anything.
- Academic wash-trade detection found stretches where up to 60% of Polymarket volume came from traders transacting across wallets they controlled, concentrated before the platform charged fees.
- Full-year 2025 sector totals range from about $44B notional in one compilation to materially different figures in others, so no honest single industry size exists.
- The practical rule for analysts is to distrust any precise volume figure whose methodology is unstated, including figures published by the platforms themselves.
Why the industry's headline number keeps moving
Every valuation, every market-share claim and every "prediction markets are bigger than sportsbooks" headline rests on a volume figure. Those figures come from venues using incompatible accounting, aggregators that revise months after the fact, and onchain records where a single trader can hold a thousand wallets. This piece sets out the four specific reasons the industry's size cannot be stated cleanly, using the measurement work in the Hodl Up Prediction Markets Report 2026. The conclusion is not that the sector is small. It is that anyone quoting a precise number is reporting a definition as though it were a fact.
The definitions do not agree, and one company proved it
Robinhood provided the cleanest illustration in a single earnings release. It reported 13.6 billion event contracts traded through its Prediction Markets Hub in Q2 2026, more than ten times the year-earlier count, and separately disclosed more than 3.5 billion contracts on Rothera. Then it warned that the two figures cannot be added, because the hub counts contracts bought or sold while Rothera counts each matched pair of yes and no contracts as one. One company, one quarter, two incompatible definitions of the same word.
The venue-level numbers carry the same problem in a less visible form.
| Metric in use | What it actually counts | Where it appears |
|---|---|---|
| Taker notional volume | Dollar value of trades hitting resting orders | The Block's venue dashboards |
| Contracts bought or sold | Unit count, each settling at up to $1 | Robinhood's Prediction Markets Hub |
| Matched yes and no pairs | One unit per matched pair, roughly half the hub basis | Rothera |
| Onchain nominal volume | Value recorded on the settlement ledger | Academic archives of the Polymarket CTF |
None of those four convert cleanly into the others. Contract counts are not notional. Onchain nominal volume is not taker volume. Researchers have also shown that a lot of Polymarket analysis double-counts because of how the contracts and event data are structured, and a 2026 academic accounting of the onchain record formalised the same problem.
The numbers get revised after everyone has quoted them
The Block first reported June 2026 combined volume at $44.8B, including an initial Kalshi print of $31.5B. Its dashboard later carried a revised June baseline of $46.95B as more trading data arrived, with Kalshi reconstructed at roughly $33.0B. The same happened to July, first published at $50.59B on 3 August and later revised to roughly $52.99B.
That matters because growth rates are ratios of two vintages. Kalshi's $37.7B July print is about 14% above the reconstructed revised-June baseline and about 20% above the initial June print. Both are arithmetically correct and they are not the same claim. August's reported 14.5% decline is measured against the revised July series, not the snapshot most coverage quoted at the time.
Up to 60% of one venue's volume was traders trading with themselves
The measurement problem may be larger than definitional inconsistency. Work led by a Columbia PhD student with Dr. Rajiv Sethi among the co-authors applied a two-part test to the Polymarket record. Wash traders show an unusually high volume-to-capital ratio, because they recycle the same capital across wallets they control. So do market makers, which is why the ratio alone proves nothing. The second test is network structure. A market maker quotes both sides and therefore transacts with a broad cross-section of the market, while a wash trader transacts systematically with counterparties that look like itself. Iterating that clustering test three or four times isolates the behaviour.
The finding was that in some periods wash trading accounted for as much as 60% of total Polymarket volume. The rate climbed around the anticipated token distribution, when traders believed volume would influence their allocation, and falls where fees apply. Sethi does not allege platform involvement, and notes that Polymarket's own data staff engaged with the findings and expressed interest in reducing the practice. His point about consequences is the sharper one, because volume claims fed the valuations and nobody audited them. The paper is posted as SSRN 5714122 and has not yet been peer reviewed.
What to do with a headline number
Three questions resolve most of the confusion before you quote a figure.
- Which definition produced it, and does the source say so?
- Which vintage is it, and does the other half of your comparison come from the same snapshot, since growth claims break silently when it does not?
- Is the venue onchain, where one trader may control many wallets and the platform itself cannot see it?
The report declines to publish a single industry size for these reasons, and treats the gaps as part of the finding rather than an omission. Our summary of the report covers the same ground for the venue-level comparison, and Sethi's full interview goes deeper on the detection method.
The full measurement chapter, including the archive-level dataset figures, is in the Prediction Markets Report 2026.
None of this is investment advice.