Hyperliquid vaults are managed accounts: a trader runs the strategy, depositors share the profit and loss. Some double in a week. The question was whether data could turn that into something systematic — and the study is a good example of why every number needs a second look.
Base rates
The public leaderboard only shows survivors, so the study included the 1,832 vaults that had closed.
All vaults, equal weight36
Exchange market-making vault158
The average vault is mostly BTC exposure with a heavy drag: a beta of 0.91 to BTC and about −5% a month on top.
Two corrections that changed the story
- Returns. With history points a week apart, a large deposit and a loss in the same step made some vaults look wiped out while they still held millions. Fixed with a standard flow-adjusted return method (modified Dietz).
- Non-investable vaults. Seven internal sub-vaults of the exchange take no deposits and produce artificial +500% weeks. One of them was a “+2,738% lottery ticket” that briefly made a rotation strategy look like ×2.7. Excluded, it was ×0.7.
What was tested
Calm vaults, only while BTC trends up+15.3%
Calm vaults, always invested−6.2%
Exchange market-making vault+17.9%
| Idea | Result |
|---|---|
| Back the top vaults by recent return | High means, negative medians, about 1 in 3 lose 30%+ the next quarter |
| Deposit after a big drawdown, withdraw after a run | Worse than depositing at random (t ≈ −7) |
| Rotate weekly into the hottest vaults | −39% to −87% a year; about 5% of picks lose half in a week |
| Calm, seasoned vaults while BTC trends up | +15% a year, −21% drawdown — still behind the exchange vault |
| Basket with profits locked into a safe vault | Locking adds 12–44% of capital, but cannot overcome a losing base |
What survived
96%of blow-ups flagged by the alarm
80%median share of the loss still ahead at the alert
42%of healthy vaults also flagged within a month
- An exit rule. A vault’s hot month that BTC does not explain reverses the next month: −6.6% on average (t ≈ −3).
- A blow-up alarm. Rebuilt from 400 blow-ups and 400 healthy vaults using the vault’s own trades: escalating activity after losses, large loss days and liquidations. It catches most blow-ups early, at the cost of frequent false alarms.