Funding Rate Arbitrage in Crypto: Delta-Neutral Spot–Perp and Perp–Perp Trades

Editorial responsibility: Oliver Sinz · Last verified: October 14, 2026

Funding rate arbitrage tries to collect perpetual futures funding payments while offsetting most of the exposure to the underlying price. The two standard constructions are a long spot position paired with a short perpetual, and a long perpetual on one market paired with a short perpetual on another. Both hedge the price level. Neither hedges the funding rate itself, the price gap between the two legs, trading costs, or the possibility that one leg is liquidated.

That makes "arbitrage" a misleading label. Nothing is locked in at entry. Funding is reset at every settlement, and a perpetual has no expiry date that forces its price back to spot; He, Manela, Ross and von Wachter stress that, unlike dated futures, perpetuals are not guaranteed to converge. The trade is better described as a hedged carry position whose result is known only after it is closed.

This guide builds the profit-and-loss ledger for both structures and works through four hypothetical BTC examples. Every price, rate, fee and return below is an illustrative assumption, not market data. Amounts are rounded to the cent, and totals are computed from unrounded values, so a column may differ from the sum of its displayed rows by a cent.

The economic idea, and the sign conventions used here

Funding is a periodic transfer between long and short holders that pulls a perpetual toward its spot index. On Binance, as on most venues, longs pay shorts when the rate is positive, shorts pay longs when it is negative, and the exchange does not keep the payment. Persistent positive funding usually signals excess demand for leveraged long exposure. Whoever supplies the short side, and neutralizes the price risk elsewhere, is paid for absorbing that imbalance.

The Bank for International Settlements' Crypto carry paper traces large, volatile crypto carry to leveraged demand from smaller trend-chasing investors and to scarce arbitrage capital, and warns that supplying the other side is risky because margin requirements can jump and positions get liquidated when prices fall sharply.

This differs from dated-futures cash-and-carry, where the futures–spot gap is scheduled to close at expiry. Funding income is a stream of variable payments with no contractual end date.

Conventions used throughout:

• Signed quantity q (in BTC): positive for long, negative for short.
• Funding rate r: the rate per settlement interval; positive means longs pay shorts.
• Reference price M: the price the venue uses to value the position at the settlement timestamp.
• Funding cash flow to a position: CF = −q × M × r. A short of 2 BTC (q = −2) with M = $60,420 and r = +0.060% receives −(−2) × 60,420 × 0.0006 = +$72.50.
• Basis: perpetual price − spot price.
• Spread (two-perpetual trades): price on the short-leg market − price on the long-leg market.

Only positions open at the settlement timestamp pay or receive. Binance allows roughly a 15-second deviation in the actual settlement time, and Bybit does not guarantee whether positions opened or closed within five seconds of the timestamp are included.

Rates, intervals, and the cash that actually moves

A quoted funding rate is a per-interval number, and intervals differ by venue and contract:

• Binance defaults to 8-hour settlement, sets intervals per contract and, since May 2, 2025, moves USDⓈ-M contracts to hourly settlement when a settlement hits the rate cap or floor.
• Bybit likewise switches a contract to hourly settlement when its rate reaches the preset limit.
• OKX supports 1-, 2-, 4- and 8-hour cycles, escalates one step at a time under stress, and divides its computed rate by 8/N so the daily-equivalent cost stays consistent across cycles.
• Hyperliquid pays every hour at one-eighth of a rate computed on an 8-hour basis.
• Kraken states that its US perpetuals settle every 8 hours, while contracts in the EU and other regions settle hourly.

Three quantities are routinely confused:

1. Per-interval rate: what the venue quotes.
2. Normalized rate: a comparison device. Converting to an 8-hour basis multiplies by 8/N, so 0.020% per 4 hours corresponds to 0.040% per 8 hours if it repeats. It is not a payment.
3. Actual cash: quantity × reference price at that settlement × that settlement's rate. Binance, Bybit and OKX use the mark price; Hyperliquid explicitly uses its oracle price instead. Inverse (coin-margined) contracts value positions as contracts ÷ price and pay funding in the coin, so all examples here use linear, USDT-settled contracts.

Rates are recomputed every interval, capped differently by venue, and can change sign from one settlement to the next. Today's rate does not determine next week's income.

Delta neutral means matched quantity, not matched notional

The hedge target is zero net exposure to the underlying, measured in BTC: q(spot) + q(perp) ≈ 0, or q(A) + q(B) ≈ 0. Contract counts matter only through contract size: 2,000 contracts of 0.001 BTC and 200 contracts of 0.01 BTC are the same 2 BTC. Dollar notionals will not match exactly because the legs trade at different prices, and sizing to equal dollar notionals would build in a small directional position.

Residual exposure creeps in through lot-size rounding, partial fills, fees or funding that accrue in a different asset than the hedge (for example, if a venue deducts a spot purchase fee from the BTC received), position reductions by the venue and, most severely, liquidation of one leg.

Structure 1: long spot, short perpetual

With equal quantity Q on both linear legs:

• Spot P&L = Q × (S₁ − S₀)
• Short perpetual P&L = −Q × (P₁ − P₀)
• Combined = Q × [(P₀ − S₀) − (P₁ − S₁)] = Q × (entry basis − exit basis)

The BTC price level cancels. What remains is the change in basis: the short perpetual gains when the premium narrows and loses when it widens. The identity excludes funding, fees, financing and any slippage not already in the fill prices, and nothing forces the exit basis to zero.

Example 1: funding captured, then partly given back

Assumptions

• Buy 2.000 BTC spot at $60,000 (cost $120,000), held outside the derivatives account.
• Short 2,000 contracts of a linear, USDT-settled BTC perpetual (0.001 BTC each, so 2.000 BTC) at $60,030. Entry basis: +$30 (+0.050%).
• Post $40,000 USDT as isolated collateral for the short.
• Funding every 8 hours (00:00, 08:00, 16:00 UTC) on quantity × mark price. Open Day 1 at 02:00 UTC, close Day 5 at 01:00 UTC: 12 settlements in 95 hours.
• Fees of 0.10% per spot fill and 0.05% per perpetual fill, paid in USDT from a separate balance; fill prices already include spread and slippage. USDT is valued at $1.
• Exit: sell spot at $61,800, buy back the perpetual at $61,890. Exit basis: +$90 (+0.146%).

The hedge ratio is 2.000 / 2.000 = 1.0 BTC. Notionals differ ($120,000 vs. $120,060) by exactly Q × entry basis = $60, which does not affect delta.

The collateral transfer is a capital allocation, not an expense. The average rate, about 0.047% per 8 hours, annualizes to roughly 51% of notional if simply multiplied by 1,095 intervals. The trade actually returned 0.126% on capital over 95 hours, about 11.6% annualized on a simple basis, and only if that stretch could be repeated all year. Costs and basis widening consumed 71% of the funding. An exit basis of $190 (about 0.31%) instead of $90 would have cut the net result to roughly $2, even though every single funding payment was positive.

The spot BTC is not collateral here, so its gains cannot support the short's margin. Under a simplified isolated model with a 0.5% maintenance rate, the short's liquidation threshold starts near $79,630 and drifts higher as funding credits accumulate. Some venues let spot holdings serve as collateral in a unified cross-margin account (Bybit's two-leg arbitrage tool runs only in cross margin), which reduces that gap but concentrates everything at one venue.

When funding turns negative

Negative funding pays longs, so the mirror trade is long perpetual, short spot. Bybit describes this as "negative arbitrage." Shorting spot means borrowing BTC through a margin or lending facility and selling it, which adds costs and constraints:

• Borrow interest, often accruing hourly (as KuCoin documents) at rates that can rise when many traders want to short. Net carry becomes funding received − borrow interest − fees ± basis change.
• Availability: borrow limits, lendable inventory, and venue or jurisdiction restrictions.
• A second liquidation regime: the short spot sits in a margin account with its own margin level and liquidation rules.

Negative funding is therefore not a symmetric opportunity, and the workbench does not model spot borrowing or short sales.

Structure 2: long one perpetual, short another

With equal quantity Q, a long on Market A and a short on Market B:

• Long A P&L = Q × (A₁ − A₀); short B P&L = −Q × (B₁ − B₀)
• Combined = Q × [(B₀ − A₀) − (B₁ − A₁)] = Q × (entry spread − exit spread)

The identity holds only while both legs are open in equal size. Each leg pays or receives funding on its own venue's schedule, rate and reference price, so quoted rates must be normalized before they are compared, and the cash must still be settled leg by leg.

Example 2: a positive differential and a losing trade

Assumptions

• Market A: long 2,000 contracts × 0.001 BTC = 2.000 BTC; linear, USDT-settled; 8-hour funding (00:00, 08:00, 16:00 UTC); fill $60,020.
• Market B: short 200 contracts × 0.01 BTC = 2.000 BTC; linear, USDT-settled; 4-hour funding (every four hours from 00:00 UTC); fill $60,080. Entry spread: +$60.
• $30,000 USDT isolated collateral on each venue; capital employed $60,000.
• Funding on each venue's own mark price; 0.05% fee per fill; fills include slippage.
• Open Day 1, 01:00 UTC; close Day 3, 01:00 UTC: 12 settlements on B, 6 on A.
• Exit: sell A at $60,850, buy back B at $60,960. Exit spread: +$110.

At entry, B quoted 0.020% per 4 hours and A 0.010% per 8 hours. Subtracting the quotes suggests a 0.010% edge, which is meaningless. Normalized, B's quote equals 0.040% per 8 hours, a 0.030% differential worth about $36 per 8 hours on $120,000 if it persisted. It did not:

Net funding was positive at every settlement, yet the strategy lost money: fees and a $50 spread widening outweighed it. The realized differential averaged about 0.023% per 8 hours, not the 0.030% implied at entry. At the realized pace of roughly $83 of net funding per day, covering fees and the spread loss needed about four days of funding in total, roughly two more days than the trade was held, and only if the spread had stopped moving. Settling on entry notional rather than each mark would have shown $163.48 of net funding instead of $165.22; in volatile markets that gap grows.

Spot–perp vs. perp–perp at a glance

Example 3: when funding fades

Take Example 1's entry, costs, collateral and 12-settlement window, but a different path: funding declines and turns negative while BTC drifts lower.

Cumulative funding peaked at $140.38 and ended at $76.83. Exiting at spot $58,700 and perpetual $58,690 (exit basis −$10) produced price P&L of 2 × ($30 − (−$10)) = +$80. Fees came to $356.12. Net: −$199.29, or −0.125% on $160,000. A perpetual discount often accompanies negative funding and helped the short here, but it is not guaranteed and was not enough.

Break-even under a constant-rate assumption

If the rate f per 8-hour settlement were constant, the settlements needed to break even are:

n* = (round-trip costs + Q × adverse basis change) ÷ (f × Q × M)

Approximating every fill and mark at $60,000, round-trip fees are 2 × (0.10% + 0.05%) × $120,000 = $360.

Rearranged, a seven-day hold (21 settlements) needs a constant rate of about 0.0143% per 8 hours to break even, or 0.0190% if the basis widens by $60. These are thresholds, not forecasts: future rates and the exit basis are unknown at entry.

Annualized funding is not an annual return

A rate of 0.010% per 8 hours multiplied by 1,095 intervals is 10.95% of notional on a simple basis. Spread over Example 1's $160,000 of capital, it is 8.21% before any cost. Earning it requires 1,095 consecutive identical settlements, no fees, an unchanged basis and no liquidation. Compounding needs a further assumption: that funding is reinvested by enlarging both legs, which means more fills and more fees. Venue displays follow their own conventions; Bybit's arbitrage documentation checked on October 14, 2026, for example, annualized the trailing three-day funding sum and halved it. Treat any such figure as backward-looking.

Example 4: isolated margin and the orphan leg

Assumptions

• Two-market structure with more leverage: long 2 BTC on Market A at $60,000 (8-hour funding); short 2 BTC on Market B at $60,060 (4-hour funding); open Day 1, 01:00 UTC.
• $12,000 USDT isolated collateral on each exchange; entry fees of $60.00 and $60.06 are taken from each account.
• Simplified margin model: equity = collateral − fees + funding + unrealized P&L at mark; maintenance margin (MM) = 0.5% × quantity × mark; liquidation when equity ≤ MM. Real venues use notional tiers and maintenance amounts, liquidation fees, partial liquidation, insurance funds and auto-deleveraging.
• Short liquidation price = (net collateral + funding credits + Q × entry price) ÷ (Q × (1 + MMR)). At entry: ($11,939.94 + $120,120) ÷ 2.01 = $65,701.46, 9.39% above B's entry. A's long threshold is $54,301.51.

BTC rallies on Day 1:

B's funding credits ($171.02) lift its threshold to $65,786.55. Shortly after 14:00 UTC, B's mark reaches it while A's mark is $65,720:

• Account B: equity $657.87, equal to MM of $657.87. The short is liquidated.
• Account A: equity $23,329.64, with $22,672.44 of buffer.
• Combined: $23,987.51 against $24,000 deposited, and net funding of +$120.66.

The portfolio was roughly flat and its funding positive, yet one account failed, because A's $11,440 unrealized gain sat at a different exchange. Moving it requires a withdrawal, a transfer and a deposit, which takes time during exactly the kind of move that causes liquidations.

Assume the liquidation process consumes B's remaining maintenance margin, so B loses its full $12,000. From that moment B earns nothing: no later funding and no recovery P&L. A is now an unhedged long of 2 BTC. It pays $45.01 of funding at 16:00 UTC (mark $64,300, rate 0.035%) and is sold at $62,400 at 18:00 UTC.

The orphan long gave back $6,640 of the price gain it held at the moment of liquidation. Had BTC kept rising, it would have made money instead; the outcome had become a directional bet. Real liquidations can be worse or different. Binance notes that a fast-moving mark price can liquidate a position beyond its displayed liquidation price, and that its process first tries to reduce a position rather than close it in full, which can leave part of a leg open. Funding debited from margin also moves thresholds.

Execution, operational and counterparty risks

• Legging risk: the two fills rarely happen at the same instant. Bybit's two-leg tool tops up a lagging leg with market orders and warns that this can cause price deviation.
• Sizing and rebalancing: lot sizes may prevent exact matching, and every collateral top-up or reduction costs time, fees and possibly a transfer.
• Settlement assets: the examples treat USDT as $1. A depeg, or legs settled in different stablecoins, changes both P&L and collateral value. Collateral eligibility and haircuts vary by venue.
• Methodology changes: intervals, caps and formulas can change; Binance and Bybit state that some interval adjustments happen without separate announcements.
• Venue risk: exchange failure, withdrawal halts or custody losses can turn a hedged position into a total loss on one leg.

Capacity and competition

Funding is paid because arbitrage capital is scarce. As more traders short the perpetual against spot, the premium and the rate tend to fall, and larger positions face wider fills and tighter borrow limits. He and coauthors find that deviations from no-arbitrage prices have diminished over time. A 2026 review by Borri, Liu, Tsyvinski and Wu reports that a Binance BTC short-perpetual/long-spot carry strategy had a Sharpe ratio of 6.45 over August 2020 to May 2025, 4.06 from 2024 onward, and a negative one in 2025, and concludes that funding premia are neither guaranteed nor permanent. Historical Sharpe ratios also say little about the jump and venue risks covered above.

What the workbench models

The Crypto Perpetuals & Basis Workbench supports hypothetical spot/linear-perpetual hedges and two-market linear perpetual funding-spread scenarios with independently modeled isolated collateral accounts. The two-market model handles different funding intervals, variable funding rates, separate market prices and path-dependent margin outcomes under defined assumptions. It does not provide live funding rates, automated execution, spot borrowing or short selling, automatic collateral transfers, cross-margin offsets, venue-specific liquidation engines or executable opportunities.

Key takeaways

• Funding arbitrage hedges the price level, not the funding rate, the basis or spread, costs or liquidation.
• Price P&L reduces to Q × (entry basis − exit basis) for spot–perp and Q × (entry spread − exit spread) for perp–perp.
• Compare rates only after normalizing intervals, and settle cash per interval on each venue's reference price.
• A positive funding differential can coexist with a losing trade once fees and spread changes are counted.
• Measure returns on all capital employed, and treat annualized funding quotes as arithmetic, not expected returns.
• Isolated accounts cannot borrow from each other; one liquidation leaves the surviving leg fully directional.