Cross Margin vs Isolated Margin in Crypto Perpetuals: Collateral and Liquidation Risk

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

Switching between cross and isolated margin does not change what you hold. Your perpetual positions, their notional value and their profit and loss are the same under either setting. What changes is which collateral stands behind which position. That decides where losses land, which buffers are available when a trade goes wrong, and whether trouble in one position can reach the others.

Below, one consistent margin model and three worked examples compare the two. All prices, funding rates and outcomes are hypothetical, not current market data or exchange-exact liquidation levels.

Key takeaways

• Isolated margin gives each position its own equity and its own maintenance test. Cross margin tests one pooled equity figure against the combined requirements of every cross position.
• Pooling can let a winning position carry a losing one. It can also let a losing position drag down positions that would have survived on their own.
• When positions lose in proportion to the margin behind them, pooling adds no extra buffer.
• Cross margin is a collateral arrangement, not a risk model. Lower requirements for hedges usually come from portfolio margin.
• Neither mode is inherently safer. Each turns the same market risk into a different pattern of liquidation risk.

Two ways to assign collateral

For linear, USDT-margined perpetuals, the two modes reduce to two simplified identities.

Isolated margin assigns a fixed amount of collateral to a single position:

• Position equity = assigned margin + unrealized P&L − costs charged to that position
• Liquidation risk arises when position equity falls to that position's maintenance margin.

Cross margin lets an eligible pool of account collateral back every cross position:

• Account equity = cross wallet balance + sum of unrealized P&L across cross positions
• Liquidation risk arises when account equity falls to the sum of all cross positions' maintenance requirements.

Binance's USDⓈ-M liquidation-price formula shows the dependency. A cross position's estimate includes the cross wallet balance plus the maintenance margin and unrealized P&L of all other contracts. In isolated mode, those terms are set to zero.

What counts as "the pool" depends on the venue and the account configuration:

• Bybit's Unified Trading Account applies one margin mode to the whole account. Cross mode converts eligible assets to a USD value and lets positions offset P&L and use unrealized profit to open new positions. Isolated mode uses only the settlement asset and allows neither.
• Binance's Multi-Assets Mode lets several coins back USDⓈ-M futures. It works only in cross margin and values BTC, ETH and BNB with a 5% haircut.
• Hyperliquid defaults to cross margin, where unrealized P&L is available as initial margin for new positions. An isolated position is liquidated using only its own margin and notional, and its unrealized P&L adds only to its own margin.

A shared pool can also shrink without any change in your positions, because collateral is revalued. In October 2025, Binance reported paying about $283 million to futures, margin and loan users who held USDe, BNSOL or WBETH as collateral during a short-lived, platform-specific depeg (The Block).

Leverage settings are not account risk

Four quantities are easy to confuse:

• Selected leverage sets how much initial margin is reserved (notional ÷ leverage).
• Position notional (quantity × mark price) generates profit and loss.
• Posted collateral absorbs the losses.
• Effective account leverage is total notional ÷ account equity.

In isolated mode, selected leverage sets the assigned margin and therefore the liquidation distance. For a linear long with a flat maintenance rate m, the simplified adverse move to liquidation is (1/L − m) ÷ (1 − m). At 5x and m = 0.5%, that is about 19.6%.

In cross mode, selected leverage mainly determines reserved margin and the balance left for new orders. Hyperliquid's liquidation documentation states that a cross position's actual liquidation price does not depend on the leverage setting. A trader can select 2x on every cross position and still run high effective leverage. Survival depends on equity relative to aggregate maintenance requirements.

The model behind the examples

Every assumption applies identically to both modes:

• Capital: 10,000 USDT in linear, USDT-margined perpetuals, with no collateral haircuts.
• Positions: 0.25 BTC-PERP at a hypothetical 100,000 USDT and 6.25 ETH-PERP at 4,000 USDT. Each has a notional value of 25,000 USDT.
• Allocation: isolated assigns 5,000 USDT to each position (5x). Cross shares the same 10,000 USDT with 5x selected, so neither mode has a free balance.
• Maintenance: a flat 0.5% of current notional per position. Cross requirements are a simple sum, with no hedge offsets.
• Prices and costs: mark price equals traded price, and moves are instantaneous. There are no fees, and funding appears only in Example 3.
• Liquidation: triggered when equity reaches maintenance margin, at position level (isolated) or account level (cross). The liquidated position or account forfeits its remaining margin. Real engines may liquidate partially, charge fees, close at a bankruptcy price or return residual collateral.

Each position starts with 125 USDT of maintenance margin. The isolated boundaries are a −19.60% move for a long and a +19.40% move for a short.

Example 1: A winning leg supports a losing leg

The account is short 0.25 BTC and long 6.25 ETH, a dollar-neutral relative-value book. BTC then rises 22% to 122,000 while ETH rises 10% to 4,400.

• BTC short P&L: −0.25 × (122,000 − 100,000) = −5,500 USDT
• ETH long P&L: 6.25 × (4,400 − 4,000) = +2,500 USDT

In cross, the ETH profit widens the BTC short's boundary. With ETH held at 4,400, solving 10,000 − 0.25 × (P − 100,000) + 2,500 = 0.005 × (0.25P + 27,500) gives P ≈ 148,706 (+48.7%). With ETH flat at 4,000, the boundary would be about 138,806 (+38.8%). In isolated mode, the ETH gain stays inside the ETH position unless the trader moves it manually, where the venue permits.

Neither outcome dominates. Cross kept the hedge but also kept the loss open, so at this instant it sits 500 USDT below the isolated account. Isolated capped the BTC loss but turned a neutral book into an unhedged 27,500 USDT ETH long. With ETH unchanged, a BTC retreat to 110,000 restores cross equity to 10,000 while the isolated account stays at 7,500. A BTC rise to 130,000 cuts cross equity to 5,000. Which structure works out better depends on a path no one knows in advance.

Example 2: Correlated losses and collateral contagion

Now the account is long 0.25 BTC and long 6.25 ETH, with the same capital and allocations.

If both fall by the same percentage x, cross liquidates when 10,000 − 50,000x = 0.005 × 50,000 × (1 − x), so x ≈ 19.60%. That is identical to each isolated boundary. Contagion appears only when losses diverge. Suppose ETH falls 2.5 times as fast as BTC:

• Isolated: ETH hits its boundary (3,216.08) when BTC is down only 7.84%. The ETH position loses its 5,000 USDT, and BTC carries on.
• Cross: solving 10,000 − 25,000b − 62,500b = 0.005 × [25,000(1 − b) + 25,000(1 − 2.5b)] gives b ≈ 11.20%. The account hits its trigger with BTC at 88,801 and ETH at 2,880 (−28.00%), where equity and requirement both equal about 201 USDT. At that point, the isolated BTC position would hold 2,200 USDT of equity against a 111 USDT requirement.

After BTC −12% (88,000) and ETH −30% (2,800):

The BTC position never breached its own boundary. It was closed because ETH's losses consumed the collateral behind it. Isolated margin kept 5,000 USDT segregated from the ETH loss. Had the cross account survived to these prices, its equity would have been −500 USDT. Full account liquidation is a model simplification; actual procedures differ.

Example 3: Funding debits and margin pressure

Return to Example 1's hedged book at entry prices. Assume a hypothetical funding rate of +0.03% per eight-hour interval on both contracts, unchanged prices, and 30 intervals (10 days). Positive funding means longs pay shorts.

• Per interval: 25,000 × 0.0003 = 7.50 USDT per position. Over 30 intervals: 225 USDT.
• Cross: the ETH long pays 225 and the BTC short receives 225. Equity and boundaries are unchanged.
• Isolated, with funding settled against each position's own margin: ETH margin falls to 4,775, moving its boundary from −19.60% to −18.69% (3,252.26). BTC margin rises to 5,225, moving its boundary from +19.40% to +20.30%.

Zero net funding can still erode the weaker leg's buffer. When both legs pay (the two longs of Example 2), the difference disappears: both modes move to an equal-move boundary of −18.69%.

Settlement order varies by venue. Bybit takes funding from available balance first. Only if that balance is insufficient does funding reduce an isolated position's margin and move its liquidation price closer. Trading fees work through the same channel; Bybit's margin calculations build the estimated fee to close into displayed initial and maintenance margin.

Hedges: economic offsets versus margin offsets

Cross margin lets profits offset losses. It does not necessarily reduce what a hedged book must post. Bybit describes both its isolated and cross requirements as calculated per individual position. Only its portfolio margin mode bases requirements on whole-portfolio risk. Binance's hedge-mode formula carries separate maintenance terms for the long and short sides of one symbol, so an exact hedge still carries two requirements.

Deribit separates two independent choices:

1. Segregated or cross collateral: where margin comes from.
2. Standard or portfolio margin: how the requirement is computed.

Standard margin sums position-level requirements and gives no hedge benefit. Portfolio margin revalues the book under price and volatility scenarios and charges for the worst result. Even Deribit's cross-collateral portfolio model does not risk-net positions in different assets, and Deribit notes that smaller futures trades can need less margin under standard margin.

Traditional "cross-margining" is closer to the portfolio idea. The OCC's program, running with CME since 1989, reduces initial margin by recognizing hedges held at different clearinghouses. That is a different concept from an exchange's cross-margin toggle.

Adding, removing and automatically topping up margin

Isolated positions can usually receive extra margin, which moves the boundary away. Removal is more restricted. Hyperliquid's "strict isolated" assets forbid it, and withdrawing unrealized P&L requires remaining margin of at least 10% of total open notional.

Bybit's Auto-Margin Replenishment, available only in isolated mode, draws on available balance as a position nears liquidation, up to the amount needed for 1x leverage. With it enabled, an isolated position can consume far more than its original allocation.

Cross positions have no per-position allocation to adjust. Protection comes from depositing collateral or reducing exposure.

How liquidation proceeds in each mode

Only the mode-dependent differences are covered here:

• Bybit: an isolated position is liquidated when mark price reaches its liquidation price, without affecting other positions. The engine cancels orders that would enlarge the position, partially closes down to a lower risk-limit tier where applicable, then closes at the bankruptcy price. Cross liquidation triggers when the account maintenance margin rate reaches 100%; the per-position liquidation price shown is only an estimate.
• Binance: at liquidation, it cancels all open orders in cross mode but only same-token orders in isolated mode. It then sends an immediate-or-cancel order and stops if margin is restored. Its Multi-Assets FAQ separately warns that once the account margin ratio reaches 100%, all positions will be liquidated.
• Hyperliquid: it first sends market orders, and leftover collateral stays with the trader. For positions above 100,000 USDC, an initial partial order covers 20%; after a partial liquidation, orders during the 30-second cooldown cover the whole remaining position. If equity falls below two-thirds of maintenance, a backstop takes over all cross positions and cross margin, but only the affected position in the isolated case.

Isolated margin limits contagion; it does not guarantee a hard loss cap. Top-ups and auto-replenishment enlarge the amount at risk. Binance warns that an insurance-fund takeover at the bankruptcy price can leave traders worse off than the displayed liquidation price implies. Its automated negative-balance clearance applies only under stated conditions.

Cross vs isolated margin compared

Trade-offs: when each structure adds risk

Isolated margin adds risk when a hedge loses one leg and leaves an orphaned directional position, when fragmented capital lets a position be liquidated while collateral sits idle elsewhere, and when automatic top-ups quietly reconnect a position to the wider account.

Cross margin adds risk when positions are correlated or one high-volatility leg dominates the losses, when volatile or haircut collateral makes the pool itself unstable, and when a low leverage setting masks high effective leverage.

The useful question is not which mode is safer. It is where losses should stop, and which buffers you are prepared to commit to which positions.

Model limits and workbench scope

The examples ignore tiered requirements, liquidation fees, slippage, insurance funds and auto-deleveraging. Real requirements rise with position size, and real liquidations are path-dependent.

The Crypto Perpetuals & Basis Workbench models generic isolated-margin scenarios only. It does not simulate cross margin, portfolio margin, automatic collateral transfers or venue-specific risk engines, so the cross-margin figures above are hand-calculated illustrations. Venue rules cited here were checked on October 14, 2026, and can change without notice.