Does Copy Trading Work? What 9,470 Real-Money Vaults Show
Copy trading can work, but rarely by following the leaderboard. We tracked every copy trading vault ever opened on Hyperliquid, 9,470 in all, with real money and real followers. 31% of followers are in profit after the leader's 10% share. The top 10 vaults by 30-day return lost a median 49% over the next year, and none of them was still in the top 10. The top 10 by balance did far better, at a median -12%. About 3 vaults in 100 were steadily ahead of a strong benchmark, and none of the ones still open is in the top 20 by recent return. Vaults with a year-long record, small drops so far and a good last six months made money 67% of the time over the following year. Switching to whoever topped the board each month did worst of all: from every start date we tested, $10,000 that followed the #1 vault was worth $25 or less a year later.
Every copy trading platform sells the same picture: find someone who trades well, press copy, and earn what they earn. The leaderboard shows the proof, a trader with +300% this month.
What the leaderboard doesn't show is what happened to the people who copied. Most platforms publish the lead trader's own return. Followers' results, and the traders who blew up and disappeared, stay out of sight.
Hyperliquid is the exception. Its copy trading runs through vaults, and every vault's balance, profit and followers are recorded on its own public blockchain, including the vaults that closed. So we downloaded all of them.
How Hyperliquid vaults work, and why the record is complete
A vault is copy trading in its simplest form. A leader opens a vault and trades it. Followers deposit into the same vault, so they own a share of one account and get exactly its result. When a follower withdraws a profit, the leader keeps 10% of it, as Hyperliquid's vault docs explain.
Because followers share the account, there is no gap between what the leader shows and what followers get, as there is on exchanges where copiers' orders fill after the leader's. And because the data lives on-chain, nothing disappears when a vault does badly. That matters: of the 9,470 vaults opened through Oct. 30, 2026, about two in three have closed, and 58% never held $1,000.
We measured every vault that ever held at least $10,000, 1,815 of them, after the profit share. Here's how they did over their whole lives.
| Vaults | |
|---|---|
Made money in dollars | 33% |
Median return over its life | -37.3% |
Median worst drop Biggest fall from a high | 61% |
Fell 90% or more at some point | 34% |
Closed | 53% |
Total won by vaults that made money | $50.2 million |
Total lost by vaults that lost money | $117.3 million |
Is copy trading profitable? What followers actually got
Vault results assume someone was in from day one. Real followers arrive and leave at different times, so we also looked at them directly: 6,078 followers with $50 or more still in a vault, 351 days in at the median.
31% of them are in profit after the profit share. The median follower is down 25.2%. Together they have put in $70.2 million, are down $9.6 million, and have paid leaders $774,000 in profit share. Holding bitcoin and ether over the same dates made money for 30% of them, so copying came out about even with holding, with much bigger swings.
- 7 to 30 days (360)38% in profit · median -1.4%
- 30 to 90 days (881)35% in profit · median -3.7%
- 90 to 180 days (736)52% in profit · median +0.4%
- 180 days to a year (1,181)44% in profit · median -1.2%
- A year or more (2,920)18% in profit · median -99.7%
The longer people stayed, the worse it got. Under a year, roughly a third to a half are in profit. Past a year, 18% are, and the median follower has lost almost everything. Counting one vote per vault, only 18% of vaults have a median follower in profit.
How does that compare with trading alone? In our study of 1,076 human traders, 31% made money and half lost 90% of their account at some point. Vault followers made money about as often, and fell 90% or more less often: 34% of vaults did. Copying didn't make people better traders. It made the worst outcome a little rarer.
What happened to the top of the leaderboard
Most new followers start with the leaderboard, sorted by recent return. So we rebuilt it for every month in the data, took the top 10 by 30-day return, and followed them forward. We did the same for the top 10 by balance, the other common sort.
| Top 10 by 30-day return | Top 10 by balance | |
|---|---|---|
30 days later | -4.0% | 0.0% |
90 days later | -19.0% | +0.3% |
180 days later | -29.4% | -1.6% |
A year later | -49.3% | -11.7% |
Lost money within a year | 74% | 60% |
Fell 50% or more within a year | 69% | 51% |
Still in the top 10 a year later | 0% | 14% |
The return leaders weren't lucky beginners. When they reached the top, they had made a median 77% in 30 days, and they had already fallen 53% from a high at some point, against 32% for everyone else. These were vaults taking big swings, and a big swing up is what puts a vault at the top of a 30-day board. The same swings came back the other way.
That matches a lab experiment published in Management Science in 2020: when people could see who had done best and copy them, they took on more risk, and the authors concluded that copy trading leads to excessive risk taking.
The board looked the same when we pulled the data on Oct. 30. The top 20 open vaults by recent return had a median worst drop of 96% and a median balance of $1,982. Sorted by balance, the top 20 had a median worst drop of 33%.
After a +40% week
A week like that is what pulls people in. We found 507 vaults with a week of +40% or more and followed each from its first such week.
| After a +40% week | After any week | |
|---|---|---|
Median result, next 6 months | -62% | -9% |
Lost money | 77% | 62% |
Vaults | 507 | 19,569 weeks |
When follower money arrives
Deposits follow performance. We found 692 weeks in which a vault's balance grew by 25% or more from new deposits. In the 90 days before, those vaults had made a median 16.7%. In the 90 days after, they lost 8.6%. An ordinary week was followed by a median 0.0%.
That's the pattern behind most copy trading losses. The money arrives after the run, and the people who join get the part that comes next.
We're now running this live: Botsfolio AI against copying the top 10 on Hyperliquid, Binance and OKX, $50,000 each, updated after every close. Watch copy trading vs AI
The steady 3%: what the vaults worth following had in common
Some vaults do hold up. To find them without hindsight, we set a test up front and ran every vault through it. A vault passes if it has a record of a year or more, ended ahead of a strong benchmark over its own dates, was ahead in at least 3 of every 4 weeks and in most 3-month stretches, and was still ahead with its single best week taken out.
The benchmark is a demanding one: Botsfolio AI's strategy, tested on real prices over each vault's exact dates. Against simply holding bitcoin and ether, 16% of vaults pass, but many of those just lost less in a falling market.
- Vaults that ever held $10,000 or more1,814 (100%)
- Ended ahead of the benchmark276 (15%)
- …with a record of a year or more217 (12%)
- …ahead in at least 3 of every 4 weeks202 (11%)
- …ahead in most 3-month stretches69 (3.8%)
- …still ahead without their single best week56 (3.1%)
56 vaults passed, 3.1% of the 1,814, or 3.6% of those with a year of history. Their median return was +188%, with a median worst drop of 36%.
What set them apart was steadiness, not size of the wins. Among the 276 vaults that ended ahead at all, the single best week made up a median 63% of the whole gain. The 56 kept winning when that week was taken out.
On Oct. 30, 25 of them were open with $10,000 or more. None was in the top 20 by recent return across all open vaults, and 5 were once vaults under $10,000 are left out. Six were in the top 20 by balance.
Copy trading strategy: split, switch or stay?
A test like that needs a year of data. A simpler version anyone can check on Hyperliquid's own pages: a record of a year or more, a worst drop under 30% so far, and up over the last six months. A single vault picked that way made money 67% of the time over the next year, against 38% for holding bitcoin and ether.
Picking is one decision. What a copier does next is the other: spread the money, move to whoever tops the board, or stay put. We tested the ways people usually play it. Each copier started with $10,000 on the first of a month, from July 2024 to October 2025, picked only from what Hyperliquid showed that day, and kept going for a year. Whenever money left a vault, the leader's 10% of the gain came out.
- Chase the #1 vault, switch monthly$0 · made money 0%
- Top 10 by 30-day return, re-picked monthly$691 · made money 0%
- Cut losers at -20%, move to the #1$5,647 · made money 12%
- Top 10 by 30-day return, never switched$6,019 · made money 12%
- 10 biggest vaults, re-picked monthly$7,122 · made money 0%
- Holding bitcoin and ether$8,682 · made money 38%
- Year-long record, small drops, up lately$12,510 · made money 67%
| How the copier played it | Made money | $10,000 became | Fell 50%+ | Beat holding |
|---|---|---|---|---|
Chase the #1, switch monthly | 0% | $0 | 100% | 0% |
Top 10 by return, re-picked monthly | 0% | $691 | 100% | 0% |
Cut losers at -20%, into the #1 | 12% | $5,647 | 75% | 6% |
Top 10 by return, never switched | 12% | $6,019 | 62% | 6% |
10 biggest, re-picked monthly | 0% | $7,122 | 56% | 31% |
Holding bitcoin and ether | 38% | $8,682 | 50% | - |
Year-long record, small drops (up to 10) | 67% | $12,510 | 0% | 83% |
What did best: a record, small drops and few switches. Up to 10 vaults passing the simple test above, re-picked monthly, made money from 67% of start dates, ended at a median $12,510, never fell 50%, and beat holding from 10 of 12 start dates. Few vaults passed early on: before September 2025, one to four a month, so for those start dates the split was really one to four vaults. From September 2025, when 5 to 22 passed, a copier who started then was at $6,582 on Oct. 29, against $6,159 for holding: a loss in a falling market, but a smaller one, with a worst drop of 43% against 59%.
Switching did the most damage. Ten vaults taken from the top of the 30-day board and left alone ended at a median $6,019. The same rule re-picked every month ended at $691 and didn't make money from a single start date. Each month's newcomers had just had their best stretch, and the profit share made it worse: a copier pays the leader 10% of every winning month on the way out, and a losing month in the next vault earns none of it back. The monthly re-picker paid a median $566 in profit share on the way to losing 93%.
Chasing the #1 vault lost almost everything. The top vault by 30-day return had made a median 263% in the 30 days before it was picked. In the month after, 9 of the 27 lost half or more, and 4 lost 90% or more. One month like that is enough, because there's little left to move: from every start date, $10,000 that followed the #1 was worth $25 or less a year later.
Cutting losers didn't fix it. Selling any vault 20% below what went in and moving the money to the current #1 ended at $5,647. The money kept landing on the next vault at its peak.
Bigger vaults were calmer. The 10 largest vaults, re-picked monthly, ended at $7,122 and fell less than the return chasers, but finished behind holding from most start dates.
Over six months, from 23 start dates, the order was much the same. Chasing the #1 ended at a median $237 and the re-picked top 10 at $4,611, while the record-based split and holding ended about level, near $10,300.
Is copy trading safe?
Copy trading on Hyperliquid is transparent: you can see a vault's full history and its followers before you deposit. The risk is in the trading. Vault leaders can use leverage and trade any market, a third of the vaults that held $10,000 fell 90% or more at some point, and a follower's money moves with every trade the leader makes.
On exchanges, two more things apply. The return a leaderboard shows is the lead trader's own account, and copiers' orders fill after the leader's, at other prices. On Nov. 2, 2026, the top 10 on Binance's 30-day leaderboard showed a median return of +2,322%. What their copiers had made on the money they put in, as Binance reports it, had a median of +9%. The top trader showed +5,709%; its 373 copiers had made $2,078 together on $64,193.
Binance, Bybit, OKX and Bitget: what carries over
The big exchanges don't publish their followers' history the way Hyperliquid does, so the full test above can't be repeated there. The parts that drove the results are the same, though:
- The leaderboard ranks by recent return. On Hyperliquid, that was the worst way to pick.
- The leader takes a share of profits. Hyperliquid's is 10%; on exchanges each lead trader sets it, and the Binance traders we looked at charged 10%.
- Copiers fill after the leader. Unlike a vault, an exchange copy is a separate account, so the copier's result is usually worse than the leader's.
We're now tracking the top 10 by 30-day return on Hyperliquid, Binance and OKX in public, day by day: copy trading vs AI, live. Bybit publishes only a short list of traders it picks itself, and Bitget publishes no leaderboard data, so they aren't in it.
Copy trading vs letting AI trade
Copy trading asks you to pick a person and trust them to keep doing what they did. The alternative is an AI that trades what it has tested, which doesn't depend on who's at the top this month. We put the two side by side here; on the same dates as the vaults above, the numbers came out like this:
| Vaults (median) | Holding BTC and ETH | Botsfolio AI | |
|---|---|---|---|
Return | -37.4% | -17.7% | +32.4% |
Worst drop | 61% | 60% | 20% |
Made money | 28% | 34% | 99% |
Over a fixed year, from the same 16 monthly start dates as the strategies above, Botsfolio AI turned $10,000 into a median $11,255 and made money from all 16. Holding bitcoin and ether ended at $8,682 and made money from 38% of them. The year-long-record split did better, at $12,510, though before September 2025 it had only one to four vaults a month to pick from.
Most of those start dates fell in a weak market for bitcoin and ether, which helped the AI. Over every 1-year window since 2018, Botsfolio AI's rules made money 80% of the time, with a median year of +32% and a worst of -49%. Holding bitcoin and ether made money 65% of the time, with a median of +28% and a worst of -82%.
Botsfolio AI trades in public, with every order on the page, and you can run it on paper money for free or with real money on your own exchange.
Is copy trading worth it? What the data says
For most people who followed the leaderboard, it wasn't. For people who looked past the leaderboard, the odds were much better. In this data, the vaults that held up had:
- A long record. A year or more of history, through more than one kind of market.
- Steady weeks. Ahead most weeks, not one huge week carrying the whole result.
- Small drops. A worst drop under 30% so far.
- A place on the balance board more often than the return board. Six of the 25 steady vaults still open were in the top 20 by balance; none was in the top 20 by recent return across all open vaults.
- Few switches. Moving to each month's leaders cost more than staying put, in losses and in profit share.
And they joined before a run, not after it. The money that arrived after a +40% week, or after a quarter of strong gains, mostly bought the fall that came next.
Method and limits
We downloaded the full vault list and every vault's history from Hyperliquid's public API on Oct. 30, 2026: 9,470 vaults ever opened (the protocol's own vault excluded), including closed ones. Returns chain each step's profit over the balance at the start of the step, so deposits and withdrawals don't count as returns; this matched 3,055 followers' recorded results (median gap 0.0, same sign 95%). The leader's 10% profit share is taken from gains. Balances already include trading fees and funding; withdrawal slippage is not modeled, which flatters copying slightly.
Follower results cover people still in a vault, up to 100 per vault, so they can lean either way: winners may have left, and wiped-out followers have no reason to. The leaderboards were rebuilt monthly from the data, not from screenshots of the live site. No vault or address is named.
The strategies start with $10,000 on the 1st of each month from July 2024. On each re-pick day they use only what the site showed then: vaults open with $10,000 or more and 30 days of history, ranked by their history up to that day, including vaults that later closed or blew up. A vault still on the list is kept as it is; money from vaults that dropped off goes in equal parts into the newcomers. The leader's share comes out of the gain since the money went in, whenever it leaves. Vault values move on Hyperliquid's weekly history points, so a fall inside a week isn't counted in the worst drop. We also divided each week's profit by the larger of the starting balance and the balance before that week's profit, so a loss on money deposited mid-week can't count against the smaller starting balance. The order of the strategies stayed the same; the re-picked top 10, for example, ended at $1,097 instead of $691.
Botsfolio AI's numbers are its strategy tested on real Coinbase prices over the same dates, spot only and with no leverage, at 0.35% per trade (a 0.10% exchange fee plus Botsfolio's 0.25%). The 1-year windows since 2018 use the same rules and fees, with a new start every week from January 2018. The vaults are real money and could use leverage and short. Holding is half bitcoin and half ether, bought once with a 0.10% fee.
FAQ
Educational analysis, not financial advice. Vault and follower results come from Hyperliquid's public data. Botsfolio AI's results are tested on past prices. Trading crypto can lose money. Past performance does not predict future results.