Anatomy of a Funding-Rate Trade: One Cycle Inside a Carry Book
The Fluxy Team
-Anatomy of a Funding-Rate Trade: One Cycle Inside a Carry Book
Our five-year study measured what funding pays in aggregate. This post is the opposite zoom level: one delta-neutral carry trade, walked end to end, with every mechanical step that the aggregate numbers hide. If you've read about "harvesting funding" and wondered what the operator actually does all day, this is that.
The setup: find the spread, size the legs
The trade: BTC funding on venue A is annualizing at 14%, on venue B at 7%. Short the perp on the expensive venue, long on the cheap one, equal notional, and collect the differential while price exposure nets to ~zero.
Sizing is where most first attempts go wrong. The constraint is not capital — it's liquidation distance on the short leg. A delta-neutral book still liquidates one leg at a time: if BTC rips 40%, your short is deep underwater on venue A while the long's profit sits on venue B, unable to help as margin. So the real sizing input is: at what price does the short leg liquidate, and how much cushion do you demand? Run at 2× effective leverage per leg and a 40% adverse move is survivable; at 5×, a normal crypto week can end the trade for you at the worst price. (This asymmetry — profits on one venue can't margin losses on another — is the whole risk of the structure, and it's why guardrails matter more here than in directional trading.)
Entry: two legs, one honest slippage bill
You cannot enter both legs simultaneously; for the seconds or minutes between fills you are directional. The discipline is to lead with the less liquid leg, size entries inside each book's depth, and accept the arithmetic: crossing the spread twice on entry plus twice on exit costs, say, 8–20bps round trip. Against a 7-point annualized spread, that's roughly a week of carry spent on admission. This is the first number that kills paper funding-arb returns, and it's knowable in advance — our backtests charge it because the venue will.
The middle: settlements pay, basis wobbles
Now the trade just… sits there. Every settlement (8-hourly on the CEXes, hourly on Hyperliquid) the short leg collects its funding and the long leg pays its cheaper funding; the differential drips into realized PnL. Meanwhile the unrealized PnL wobbles — sometimes alarmingly — because the two perps' prices don't move in lockstep. This basis noise is the psychological test of the trade: the book shows −2% one day and +1.5% the next while the realized carry ticks up in the background. Operators who don't understand that the wobble is mark-to-market noise on a hedged structure close good trades at the worst moments. (Operators who over-trust it forget the wobble becomes real if a leg liquidates. Both errors are fatal; only one is common.)
The other mid-life event: the spread moves. Funding differentials mean-revert and flip. A trade entered at 7 points of spread might compress to 2 within weeks — at which point the carry no longer pays for the risk, and the discipline is a rules-based exit, not hope.
Exit: the same slippage, plus the verdict
Unwind both legs (again: less liquid first), pay the spread again, and tally: settlements collected, minus four spread-crossings, minus fees, plus/minus whatever basis existed between entry and exit marks. On a well-run cycle the settlements dominate; on a badly timed one the basis move eats the quarter's carry. The honest expected value comes from running this loop across history — which is exactly what a settlement-clock backtest does, and what a rate-times-notional estimate does not.
Why this is an infrastructure trade
Notice what the edge consisted of: settlement-accurate data to find real spreads, sizing math driven by liquidation distance, execution across two venues, and the temperament to hold through basis noise — nothing exotic, everything operational. That's the general truth about carry: it pays operators, not spectators. The whole loop — the data, the funding-aware backtest, the multi-venue execution with reconciliation guards — is what the platform runs, because it's the loop we run ourselves.
Backtest a carry book against settlement-level funding history, then run both legs from one portfolio. Start with the funding-arb template →
Related reading