Graveyard / Copy trading / Hyperliquid
Copying Hyperliquid vaults
Does picking Hyperliquid vaults by track record work?
The idea
Hyperliquid vaults and well-known accounts show their history in public. Pick those with a good track record, copy their trades a minute behind, and let their skill do the work.
How we tested it
| Data | 58 usable vaults and 31 public trader accounts. Weekly rolls on 90-day trailing evidence, copy 60 seconds late, 5% stop, exit when the trader exits. Also variants on multi-window consistency. |
|---|---|
| Period | Eight weekly rolls, 24 August to 20 October 2026. |
| Costs | 4.5 bp taker plus 1.5 bp slippage per side, and a 1.5 times stress case. |
| Holdout | Two halves reported separately. |
| Control | The unselected pool of vaults. A separate study of depositing into vaults chosen by trailing share-price statistics (1,710 vaults, 22 monthly rolls, 10% profit share) reported: primary rule −0.84% per 30 days, random-pick percentile 0.52, and the vault run by the exchange itself (HLP) at +1.09% per 30 days (t 2.2). |
Result
| Selected vaults, copies | 2,329 |
|---|---|
| Per copy / profit factor | −0.17% / 0.80 |
| Halves | +0.00% / −0.28% |
| Unselected vaults | −0.20%, profit factor 0.87 |
| Public trader accounts, 1,135 copies | −0.22%, profit factor 0.86 |
| Multi-window, 1,996 copies | −0.18%, both halves negative |
Why it failed
Selection added nothing over the unselected pool. In the deposit study the edge over random flipped sign between the halves for every rule, and the first half's edge came mostly from avoiding blow-ups that dominated 2025. The exchange's own market-making vault beat every rule in both halves, but that is a passive yield with a market-maker loss tail, not a copy edge.
What would make it worth retesting
No for selecting by track record. The exchange's own vault is a risk asset, not a trading edge.
Test id: hlsrc. Past results of a backtest say nothing certain about the future.