We classified every Polymarket wallet by how it trades across two quarters of 2026. Liquidity on a venue with close to a million active wallets comes from a few dozen addresses, retail loses 14.5% on what it holds to resolution, and the only cohort that finishes positive is 271 wallets doing 5.4% of aggressive flow.
We analysed wallet-level activity on Polymarket’s international venue across March–May and June–August 2026 to understand who supplies liquidity, who takes it, who holds risk, which markets they trade, and which groups contracted as volume fell. We classified wallets using three signals: whether they make or take liquidity, how often they trade both sides of a market, and how many trades they make each month. The thresholds are arbitrary, so we varied each one and reran the analysis to test how sensitive the results were. Market makers and semi-professional wallets supplied 34.8% of resting orders under every specification. The automation share was more sensitive to the trade-count threshold, ranging from 29.7% to 43.7%; we report it as a range throughout.
All figures cover 1 March – 31 August 2026, with market resolutions tracked to 1 September. Volume means taker-side notional, liquidity supplied means maker-side notional
Passive accumulators supply 27.3% of resting flow against market makers' 18.7%. They work a far wider book — the median accumulator posts in 2,508 markets against a market maker's 180 — almost never take both sides, and end up holding what fills, at a net-to-gross of 0.84. Dedicated liquidity, market makers and semi-professional wallets together, supplies 35% in summer against 29.8% in spring, the increase reflecting retail flow leaving rather than more capital arriving.
The taker side is the mirror image: retail takes 30.5% while market makers take 1.7%. Between quarters bots and arbitrage took 37.5% of aggressive flow against 28.7% in spring, and retail fell from 42.5% to 35.1%.
Positions held through to resolution return +1.8% for professional directional wallets and −14.5% for active retail, which pays 40.1 cents a share and receives 34.3. Long-tail retail loses 10.2%. Every liquidity-providing group lands within half a point of zero: market makers −0.1%, semi-professional −0.2%, pair-arbitrage −0.3%, passive accumulators −0.5%. That is what a functioning market-making book should look like, since the inventory is a by-product and the earnings sit in the spread.
Retail wins more often than professionals — 53.8% of its positions against 44.7% — and still loses money. Its winners are small and its losers are large: only 34.2% of the shares it holds resolve in the money, against 54.7% for professional directional.
Retail also exits early, selling before resolution on 46.2% of its taker notional against 8% for professional directional, and that is excluded by the numbers above.
Polymarket had 966,750 active wallets in the summer quarter, but a disproportionate share of the flow runs through very few of them. Thirty addresses supply 21.7% of all resting liquidity on the venue, up from 14.9% in spring, and ten pair-arbitrage addresses take 8.4% of all aggressive flow, up from 3.4%.
Concentration increased on both sides of the market as the venue shrank. Retail and directional traders are meaningful cohorts. Liquidity and arbitrage, by contrast, amount to a small number of individual addresses and are better considered case by case than as trader categories.
Polymarket traded roughly $25bn of taker-side notional per quarter against $366–414m of open positions. That gap between trading and holding is not evenly distributed. Active retail held 33.6% of open interest while accounting for 13.2% of liquidity supplied. Market makers were at the other end: they supplied 18.7% of liquidity, accounted for 1.7% of taker flow, and held 1.9% of capital. Directional traders held 1.5–2.3 times their share of flow. The same pattern appeared in both quarters.
Wallets were classified from their trading behaviour alone. Position balances were not used to define the groups, so the differences in capital held reflect how these groups actually traded.
Taker notional fell 13% between quarters. Sports was the only major category to grow, up 34%, driven entirely by the 2026 FIFA World Cup, which added $7.4bn. Excluding the tournament, sports fell 34% and the venue overall fell about 41%.
The shift in sports also compressed the differences between trader groups. In March–May, market makers and pair-arbitrage put ~46% of their flow into crypto and largely avoided politics, while professional directional traders put just 3.1% into crypto and 85% across sports and politics. The gap was 37 points. By June–August, it was 20 points, with every group putting 69–89% of flow into sports.
The contraction was concentrated in retail. Retail dollars fell 28%, while automation rose 4% and directional trading 5%. Active wallets fell 37–42% across automation, liquidity and retail, while directional wallets fell 15%.
Polymarket's international liquidity remains concentrated. Thirty addresses supply a fifth of resting liquidity and ten take over 8% of aggressive flow, so headline volume describes a narrower set of participants than it appears to. Institutional participation looks limited here, though this is the offshore book and US persons cannot access it.
For price discovery, we would count taker flow from professional directional wallets and exclude maker and arbitrage flow, which is largely recycled, as well as near-certain strikes in ladder-heavy categories such as finance and rates. The first filter leaves 5.4% of taker notional, from 271 wallets; the second takes finance from $587m to $189m.
Finance is the cautionary case for anyone looking for institutional activity. It is 2.5% of platform notional and shrinking, down 47% between quarters; two-thirds of it is multi-outcome rate ladders, where most strikes sit near a dollar so a large order buys less exposure than large orders anywhere else; and 21 to 35 wallets trade it near-exclusively, with five addresses supplying most of the large-ticket money in each quarter.
Prediction markets are event contracts, and a few large events can reshape the market mix quickly. Participant types were relatively stable across six months, but the markets they traded were not: the World Cup accounts for the entire increase in sports, and several patterns seen in June–August were absent in March–May. We would therefore look across a longer horizon before drawing conclusions about how these participants behave.
Also notable these weeks
CLARITY Act fails On 15 September the Senate refused cloture on the Digital Asset Market Clarity Act, 49–50, short of the 60 votes needed to take the bill up. That left no statute splitting SEC/CFTC jurisdiction or giving crypto firms a durable rulebook. Both chairs said they would not wait and would act under existing authority.
Congress did not pass market structure. The agencies still opened two doors so firms can build inside a temporary legal box instead of waiting for the next vote.
Looking to use Dune for your company?
