# 1.6% of Polymarket Wallets Never Sleep. They Made 58% of All Trades.
Hourly activity for 130,346 Polymarket wallets over 12 days in August and September 2026: 2,080 wallets were active in every hour of the clock and filled 58.4% of 9.82 million trades. The ten busiest wallets alone filled one trade in five. Half of all wallets traded on a single day.
Canonical URL: https://vultax.com/research/polymarket-wallets-that-never-sleep-trading-cadence
Author: [Vultax Research](https://vultax.com/editorial-policy)
Published: 2026-09-09
Last public revision: 2026-09-09
Published research snapshots and calculations; these exports do not contain current quotes or live account data. Observation windows and populations are stated in the source captions and methodology. Missing metadata is unknown, not zero. Figures describe trading cadence observed by Vultax's prediction-market pipeline over the stated window, aggregated from public on-chain fills into hourly buckets. A trading pattern is not proof of automation, and nothing here identifies any wallet's operator, alleges wrongdoing, or constitutes financial advice.
## Key figures
- Wallets active in every hour of the clock: 1.6% — 2,080 of 130,346 wallets seen between 22 August and 2 September 2026. Together they filled 58.4% of 9.82M trades and 26.9% of $448M notional
- Trades filled by the ten busiest wallets: 20.9% — The busiest single wallet filled 947,876 trades in 12 days, roughly one a second. All ten were active on all 12 days and in all 24 clock-hours
- Wallets that traded in a single hour: 42.6% — 55,548 wallets. Half of all wallets (50.1%) were active on one day only; the median wallet made 3 trades worth $40
- Median trade, round-the-clock wallets: $7.50 — Against $8.67 for everyone else. The cohort's median wallet filled 498 trades at 5.4 per active hour; its 90th percentile filled 4,335
## What was measured
Vultax's prediction-market pipeline materialises Polymarket activity into hourly buckets: for every wallet, every market and every UTC hour, the number of fills, the buy and sell split and the notional traded. This study reads those buckets for the 12 complete days from 22 August to 2 September 2026. In that window the pipeline saw 130,346 distinct wallets, 9,816,634 trades and $448.5 million of notional across every market it tracks.
The question is cadence, not identity: when does each wallet trade, and how much of the market do the wallets that trade around the clock account for? A wallet is called round-the-clock here if it was active in all 24 hours of the UTC day at some point during the window. That is a description of behaviour. It is consistent with automation, and it is also consistent with a desk that never closes; the study does not claim to know which, and it never names a wallet.
One coverage note that matters for reading the numbers. The pipeline publishes hourly buckets as capacity allows, and over these 12 days it materialised 153 of the 288 possible hour-buckets. Every hour of the clock was observed at least twice, most of them five to ten times. Wallet counts and trade totals are exact for the buckets observed; the hour-of-day profiles below are normalised per observed bucket so that a well-covered hour does not look busier than it is.
## Who never sleeps
2,080 wallets, 1.6% of the population, were active in every hour of the clock. They filled 58.4% of all trades and 26.9% of all notional. Their share of trades was stable day to day, between 54.4% and 61.7% on every one of the 12 days, which is the signature of a standing presence rather than a burst.
The distribution inside the cohort is steep. Its median wallet filled 498 trades over the window, at 5.4 trades per active hour and $7.50 per trade; its 90th percentile filled 4,335. At the very top, the ten busiest wallets on Polymarket filled 20.9% of every trade in the window between them. The busiest filled 947,876, which is one trade every 1.1 seconds for 12 days without a pause; the second filled 478,317. All ten were active on all 12 days and in all 24 clock-hours. The top 100 wallets by trade count filled 38.9% of trades and the top 1,000 filled 62.4%.
Widen the definition and the picture holds. Wallets seen in at least 22 of the 24 clock-hours were 3.1% of the population and 67.9% of trades. Wallets active on all 12 days were 2.8% and 57.5%. Wallets active in at least three quarters of the observed hours on five or more days were 1.1% and 56.7%. However the cut is drawn, a population of one to three thousand wallets is on the other side of most fills.
### Who never sleeps
Polymarket, 22 August to 2 September 2026, 130,346 wallets, 9,816,634 trades. Cuts overlap. Shares of trades and notional are of the window totals.
| Cut | Wallets | Share of wallets | Share of trades | Share of notional |
| --- | --- | --- | --- | --- |
| Active in all 24 clock-hours | 2,080 | 1.6% | 58.4% | 26.9% |
| Active in 22+ clock-hours | 4,007 | 3.1% | 67.9% | 35.3% |
| Active on all 12 days | 3,625 | 2.8% | 57.5% | 36.6% |
| All 24 clock-hours and all 12 days | 1,341 | 1.0% | 50.9% | 25.3% |
| Active in 75%+ of observed hours on 5+ days | 1,451 | 1.1% | 56.7% | 27.0% |
| Ten busiest wallets by trade count | 10 | 0.008% | 20.9% | 5.0% |
| Traded in a single hour | 55,548 | 42.6% | 1.1% | 2.1% |
| Traded on a single day | 65,362 | 50.1% | 2.7% | 4.3% |
[CSV](https://vultax.com/research/polymarket-wallets-that-never-sleep-trading-cadence/data.csv?dataset=section-who-never-sleeps-table) · [JSON](https://vultax.com/research/polymarket-wallets-that-never-sleep-trading-cadence/data.json?dataset=section-who-never-sleeps-table)
## They keep the same hours as everyone else
The expectation was that round-the-clock wallets would trade flat across the day and the rest of the market would show a sleep trough. The data says something more useful. Trade share by hour of day is almost identical for the two groups: the round-the-clock cohort peaks at 13:00 UTC and troughs at 07:00 with a peak-to-trough ratio of 1.86; everyone else peaks at 15:00, troughs at 03:00, ratio 1.88. By six-hour block the cohort's trades fall 22% in 00:00 to 06:00, 22% in 06:00 to 12:00, 31% in 12:00 to 18:00 and 25% in 18:00 to 24:00; the rest of the market is 20%, 23%, 32% and 25%.
The always-on wallets are present at 04:00, but they trade most when the crowd trades most. That is consistent with behaviour that scales with incoming flow, the way a market maker or a fill-on-signal system does, rather than with independent overnight conviction. At $7.50 a trade, 498 trades per wallet and one fill a second at the top, the cohort looks like the other side of the market's order flow, not a set of large directional bets. The [Vultax fair-value study](https://vultax.com/research/prediction-market-fair-value-calibration) found systematic distortions in Polymarket prices; the wallets described here are the population most likely to be harvesting them at scale.
### Share of trades by hour of day, UTC
22 August to 2 September 2026, normalised per observed hour-bucket. Both groups peak in the US morning.
| Category | Round-the-clock wallets (%) | Everyone else (%) |
| --- | --- | --- |
| 00 | 3.7 | 3.3 |
| 01 | 4 | 3.5 |
| 02 | 3.9 | 3.6 |
| 03 | 3.6 | 3 |
| 04 | 3.4 | 3.1 |
| 05 | 3.6 | 3.1 |
| 06 | 3.4 | 3.4 |
| 07 | 3.1 | 3.1 |
| 08 | 3.6 | 3.8 |
| 09 | 3.5 | 3.6 |
| 10 | 3.9 | 4.2 |
| 11 | 4.3 | 4.6 |
| 12 | 4.6 | 5.1 |
| 13 | 5.8 | 5.6 |
| 14 | 5.6 | 5.6 |
| 15 | 5 | 5.6 |
| 16 | 5.1 | 5.4 |
| 17 | 5 | 5.2 |
| 18 | 5 | 5.2 |
| 19 | 4.7 | 5 |
| 20 | 4.3 | 4.6 |
| 21 | 3.9 | 3.8 |
| 22 | 3.9 | 3.6 |
| 23 | 3.3 | 3.1 |
[CSV](https://vultax.com/research/polymarket-wallets-that-never-sleep-trading-cadence/data.csv?dataset=section-they-keep-the-same-hours-as-everyone-else-chart) · [JSON](https://vultax.com/research/polymarket-wallets-that-never-sleep-trading-cadence/data.json?dataset=section-they-keep-the-same-hours-as-everyone-else-chart)
## Most wallets are visitors
The other half of the distribution is the mirror image. 42.6% of wallets traded in exactly one hour of the 12 days, and 50.1% on exactly one day. The median wallet across the whole population filled 3 trades worth $40 and was active in two hours. Wallets that were not round-the-clock had a median trade of $8.67, slightly larger than the cohort's $7.50, and a median notional of $40 against the cohort's $4,629.
This is the same shape the [P&L concentration study](https://vultax.com/research/best-polymarket-traders-leaderboards) found from the other direction: over 18 June to 14 July, 54% of wallets closed nothing at all and the top 1% of winners took 44% of realised gains. Polymarket's active population is a small standing core and a very large transient edge, and the core does the trading.
### Wallets by number of days active
130,346 wallets over 12 days. Half were active on one day only; 3,625 on all twelve.
| Category | Wallets |
| --- | --- |
| 1 | 65362 |
| 2 | 17532 |
| 3 | 10059 |
| 4 | 8395 |
| 5 | 6474 |
| 6 | 4586 |
| 7 | 3794 |
| 8 | 3196 |
| 9 | 2591 |
| 10 | 2381 |
| 11 | 2351 |
| 12 | 3625 |
[CSV](https://vultax.com/research/polymarket-wallets-that-never-sleep-trading-cadence/data.csv?dataset=section-most-wallets-are-visitors-chart) · [JSON](https://vultax.com/research/polymarket-wallets-that-never-sleep-trading-cadence/data.json?dataset=section-most-wallets-are-visitors-chart)
## What this says about the "30% are agents" figure
The statistic in circulation this year is that more than 30% of Polymarket wallets are AI agents, with 14 of the top 20 profitable wallets described as bots and agents reported to end positive 37% of the time against 7 to 13% for humans. Coverage traced through [Turbine's May 2026 summary](https://www.turbinefi.com/blog/how-to-compete-with-ai-agents-prediction-markets-2026) attributes the first to CoinDesk (15 March 2026), the second to Finance Magnates and the third to Bloomberg (28 April 2026). Vultax has not verified any of the three and the original CoinDesk page refused automated reading on 5 September 2026.
Cadence cannot identify an agent, so this study neither confirms nor refutes those numbers. It bounds them. If 30% of wallets are agents, then the great majority of those agents behave like the visitors above, trading a few times and leaving, because the population that behaves like always-on automation is 1.6% by the strict cut and 3.1% by the loose one. And whatever the count, the wallets that never sleep are where the volume is: two thirds of every fill on Polymarket, day after day, comes from one to four thousand addresses. Self-described operators exist and publish their own figures, [Olas's Polystrat](https://olas.network/blog/introducing-polystrat-an-autonomous-ai-prediction-agent-on-polymarket) among them; none of them is identified here.
## What it means if you trade by hand
The practical reading of a market where 1.6% of participants fill 58% of trades.
- Your counterparty is probably always on - price a fill as if the other side refreshes every second, because on the busiest markets it does.
- Post, do not take - Polymarket charges takers C × feeRate × p × (1 − p) and "Makers are never charged fees" ([fee docs](https://docs.polymarket.com/trading/fees.md)); against a standing counterparty, a resting limit order is the only way to keep the 4 to 7% of expected value the fee would take.
- Do not read speed as conviction - a wallet that fills one trade a second is expressing a process, not a view; the rare informative print is the large trade from a quiet wallet, which is why [whale trackers](https://vultax.com/research/best-polymarket-tools-2026) that score on wallet history beat those that score on size.
- Copying an always-on wallet is copying its costs - detection lag, the taker fee and the fill it just moved land on you at every one of its hundreds of trades; the [copy-trading arithmetic](https://vultax.com/research/polymarket-copy-trading-costs-and-tools) is on a separate page.
- Test against live prices first - the Vultax paper-trading desk fills simulated orders against the live book with no money at risk.
## How to read the live figures
The strip at the top of this page is rewritten every ten minutes from Polymarket's public leaderboard by day, week and month, its data API's trades of $10,000 or more, its global open interest, and Vultax's record of every fill it observes, including the always-on wallet cohort. The Vi IQ beneath it scores the six domains the Vultax terminal scores for a pair, computed for this page's subject from published inputs, with a domain that cannot be computed shown as unavailable rather than filled in. The figures in the body of the article are the measurement it was written on and are dated; the strip is what the same measurement says now.
The first two figures are the always-on cohort's share of the wallets that traded in the last day and of the dollars they traded: a small share of wallets, a large share of money, which is the article's finding measured fresh every ten minutes. The third is how many of the last 24 hours had any fill at all, and the fourth the number of wallets Vultax saw trade.
## Context from elsewhere
The 'never sleeps' cohort has a candidate explanation from outside this site. Columbia researchers flagged roughly one Polymarket wallet in seven as wash trading by counterparty structure, with many flagged wallets making no profit, and found the flagged share of volume peaking near 60% in December 2024 before falling and returning to around 20%. Round-the-clock trading with no profit is what incentive-farming looks like; round-the-clock trading with profit is what a market maker looks like. The two are separable in the data this article uses, and the profit concentration page on this site does the separation.
The population is changing around them. Polymarket's U.S. exchange came off its waitlist in May 2026 and its perpetual futures launched in September; the CFTC's February 2026 advisory names wash sales, pre-arranged trading and disruptive trading as conduct it polices on any designated contract market, which reaches automated accounts on U.S.-regulated venues directly. Whether the always-on share rises or falls from here is exactly what the strip on this page will record.
## Limits of this measurement
Twelve days is a short window and the last week of August is not every week. The pipeline materialised 53% of the possible hour-buckets, so the absolute trade counts undercount the window and the clock-hour cut is stricter than it looks: a wallet had to appear in every hour that was observed. Cadence describes behaviour, not identity; a wallet that trades every hour may be an agent, a desk, or a shared execution service used by many people. Notional is computed from fills and does not reflect open positions. The study covers Polymarket only and does not link any wallet to a person. The table behind every figure is offered as a CSV below.
## Questions
### How many Polymarket traders are bots?
Nobody knows exactly. In this sample 1.6% of wallets traded around the clock and made 58% of trades; the live strip shows the cohort's share of wallets and dollars in the last day. Trading in every hour of the day is the observable property; whether a program does it is an inference.
### What is an always-on wallet?
One that traded in enough distinct hours across the day, over a long enough window, that no person keeping any schedule would produce the pattern. The method section gives the thresholds.
### Are always-on wallets wash traders?
Some may be: a 2025 Columbia study flagged about one wallet in seven as wash trading, often with no profit. Others are market makers earning the spread. The profit-concentration page on this site separates the two.
### What does this mean for trading by hand?
That most of the counterparties in any market at any hour are programs, and that the hours a person keeps are not the hours the market keeps. The article's section on trading by hand covers how to use that.
## Sources and methodology
- [Polymarket trading fees](https://docs.polymarket.com/trading/fees.md) — "Makers are never charged fees. Only takers pay fees." Read 2026-09-05
- [Turbine: How to compete with AI agents in prediction markets (2026)](https://www.turbinefi.com/blog/how-to-compete-with-ai-agents-prediction-markets-2026) — Attribution chain for the agent-share statistics; 21 May 2026
- [Olas: Introducing Polystrat](https://olas.network/blog/introducing-polystrat-an-autonomous-ai-prediction-agent-on-polymarket) — A self-described autonomous agent operator, 9 February 2026
- [Vultax: Where Polymarket Profits Go](https://vultax.com/research/best-polymarket-traders-leaderboards) — The same population seen through realised P&L
- [Vultax: Do Profitable Polymarket Traders Stay Profitable?](https://vultax.com/research/best-polymarket-traders-leaderboards) — Activity as the first filter on persistence
- [Vultax: Polymarket prices tested against 6,776 resolved markets](https://vultax.com/research/prediction-market-fair-value-calibration) — The distortions an always-on counterparty can harvest
- [Vultax prediction desk](https://vultax.com/features/prediction-desk) — Where the hourly activity buckets behind this study are rendered
- [Gizmodo — Study finds around a quarter of Polymarket trades are fake (Columbia, Nov 2025)](https://gizmodo.com/study-finds-around-a-quarter-of-polymarket-trades-are-fake-2000683231) — About 25% of Polymarket volume over three years flagged as wash trading by counterparty structure, peaking near 60% in December 2024; roughly one wallet in seven flagged; many flagged wallets made no profit.
- [Finance Magnates — Polymarket rolls out U.S. app after CFTC green light, starting with sports](https://www.financemagnates.com/forex/polymarket-rolls-out-us-mobile-app-after-cftc-green-light-starting-with-sports-events/) — Polymarket US launched 2 December 2025 on the QCEX designation acquired for $112 million; the invite waitlist was removed in May 2026; iOS only at the time of writing.
- [Coinpaprika — Polymarket expands beyond prediction markets with 20x leverage perps (Sep 2026)](https://coinpaprika.com/news/polymarket-expands-beyond-prediction-markets/) — Perpetual futures launched 3 September 2026 with up to 20x leverage; 67 markets within hours; U.S. traders barred. Kalshi's bitcoin perpetuals went live 3 June 2026.
- [CFTC — Enforcement Division issues Prediction Markets Advisory (25 Feb 2026)](https://www.cftc.gov/PressRoom/PressReleases/9185-26) — Insider trading, fraud and manipulation, wash sales and pre-arranged trading, and disruptive trading named as conduct the Commission polices on any designated contract market; cites a candidate trading his own race and a video editor trading on non-public channel information.
Changing market context is available on the [article page](https://vultax.com/research/polymarket-wallets-that-never-sleep-trading-cadence). It is separate from the published study exported here.