Draw the line first: the “risk-free” in “risk-free arbitrage” means no directional price risk. It does not mean no risk.
When those two get run together, a very specific technical claim gets read as a blanket promise. Below are the four risks one at a time: which one is cancelled, which three are not, and who each one lands on.
The one that is cancelled: price direction
The two legs face opposite directions with matched notionals, so their P&L mirrors each other. Bitcoin can drop thirty percent in a day and the combined position barely moves — it was never betting on direction.
What cancels it is not a forecast. It is the two notionals actually matching. Spot and perpetual are two markets with their own lot sizes and tick sizes, so a quantity derived by dividing notional by price gets truncated in different directions on each side: you think you hedged 100% and the real figure is 99.7%, with 0.3% sitting naked. Once is nothing. A few hundred times and the drift becomes a real directional position.
The gap between the two notionals is the opening residual. It is not zero, only small. It belongs in its own column, quoted in money rather than as a percentage; how both legs are landed together and how a single-leg fill is rolled back is the whole subject of what makes delta neutral hard in practice.
Still there: funding thins out, or goes negative
Funding is set by market sentiment. In flat and bear markets it drifts toward zero, and in stretches it turns negative. This one needs no imagination — it is sitting on our year-by-year track record:
| Year | ETHUSDT raw | With scaled capital |
|---|---|---|
| 2021 | 37.562% | 375.621% |
| 2022 | 0.787% | 7.873% |
Same instrument, one year apart, a factor of nearly 48. BTCUSDT in 2022 managed only 4.165% (41.649% scaled). For contrast, across 2021–2025 — the five complete years where every settlement can be checked — the ten cells average 12.43% raw, 124.296% scaled. The average and the worst cell are more than ten times apart, which is why an average on its own tells you nothing.
Negative is not a hypothetical either. In the same snapshot (as of 2026-08-25), SOLUSDT was at −1.126% for 711 settlements year to date — −11.26% scaled. Holding that leg for the year did not just earn nothing, it paid out. What the flipped sign means, and whether it can be traded the other way round, is in negative funding rates.
State the nature of this one precisely: the return thins out and can go negative, but the principal is not harmed by price direction. It is not a risk that blows up the account. It is a risk that leaves you holding a position for nothing.
Still there: the exchange and extreme markets
Exchange downtime, contract rule changes, liquidity drying up in a wick — none of that is inside the set of things a strategy can cancel.
Rule changes are the ones most often filed under “won’t happen”. From 18 September 2025 Binance USDT-margined perpetuals added an 8 / N frequency normalisation factor to the formula and moved a batch of contracts from 8-hour settlement to 4-hour and even 1-hour. Run a 4-hour contract through the old formula and the figure comes out exactly double — and nothing will raise an error on your spreadsheet when that happens.
Delisting is the second underrated one: the top rows of any funding leaderboard are often contracts on their way out. Whether a row is workable comes from dividing round-trip cost by the per-settlement rate. On our own accounts we measured INJUSDT at 0.30% round-trip fees plus 0.048% round-trip spread — a row advertising “40% annualised” needs roughly twenty settlements to break even, and funding often flips sign within two. Zero collateral ratio and empty borrow inventory are the same kind of trap, listed one by one in six hidden costs.
What the system can do is set thresholds, auto-deleverage and alert. What it cannot do is stand behind the exchange. Anyone claiming otherwise has just handed you a risk signal.
Still there: the other side of scaled capital
This scheme scales capital by 10x, and funding is charged on notional: notional = mark price × position size. 1,000 USDT at 10x is 10,000 USDT of notional, so a 0.01% settlement is 1 USDT, not 0.1 USDT.
The same multiple that enlarges the income enlarges the costs and the swings: what you pay out during negative stretches also runs at 10x, and fees and spread are charged on notional too. As margin ratio approaches the maintenance line the system deleverages automatically, and deleveraging itself thins the period’s return.
The one custody adds: platform credit risk
The four above belong to the strategy. This one only appears once you choose custody.
Funds held in the platform’s dedicated account and traded by the platform is what makes a floor or a principal guarantee performable — and the price is that the money sits in someone else’s account. Custody is not principal protection, and transparency does not cancel credit risk.
What can be done is to put the verifiable part on the table. Six items are available to an investor in real time: balance, per-trade history, positions, the return curve, subscription records and withdrawal status — the full custody and verification basis is set out on that page. The solvency ratio is not among those six — it sits behind admin permissions and an investor cannot query it. Describing something unqueryable as queryable is worse than leaving it off.
Who each risk lands on
Of the categories above, exchange risk and extreme markets are backstopped by nobody. Who eats the loss when the strategy underperforms depends on which plan you hold. All three carry the same 2% annual management fee, taken out of the return — in a thin year the platform collects less and never bills the investor for the shortfall. The difference is protection and split:
| Plan | 100% principal guarantee | Floor APR | Excess split (platform : you) | Return risk |
|---|---|---|---|---|
| Capital and yield protected | yes | 7% (take-home) | 100 : 0 | platform carries it |
| Capital protected, 7:3 | yes | none | 70 : 30 | moves with the strategy |
| No capital guarantee, 3:7 | no | none | 30 : 70 | principal and return can both fall |
The split worth stating out loud: principal protection is not yield protection. The middle plan returns every unit of principal, but what it earns that year depends on the strategy. All three start at 5,000 USDT, with terms of one to five years by tier, and the full side-by-side is on plans and pricing.
One more thing belongs up front: every “yes” and every “platform carries it” in that table is performed by the platform itself, which means all of them sit on top of the credit risk from the previous section. Leave that layer out and the table looks better, but it is not what is happening.
Turning risk into something comparable
Take any arbitrage offer and split its risk description into three questions:
- Which category is cancelled, and by what mechanism?
- Which ones are not cancelled, and are they on the same page?
- For the ones that are not cancelled, who eats the loss?
A document that answers only the first is not a risk disclosure. It is a sales sheet.
Turning those three questions into a yardstick for a specific platform is four questions to settle first; how these risks end up as the number on your own statement is what a calculator leaves out.