Crypto Futures Liquidation Explained: Maintenance Margin, Mark Price and Leverage

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

In crypto derivatives, liquidation is the forced reduction or closure of a leveraged position by the venue's risk engine once the collateral behind it no longer meets the venue's minimum requirement. It is not the same as "liquidating" coins you own, which simply means selling them: a spot holding bought outright, with no borrowed exposure, has no margin requirement to fail. And unlike a voluntary exit, where the trader sets the exit rule, a liquidation leaves the moment, the method and often the price to the venue.

This guide explains how that moment is determined, how leverage, the mark price, fees and funding shape it, and why outcomes on a real exchange can differ from any formula. Every price, rate and position below is a hypothetical illustration, not market data.

What liquidation is, and why venues force it

A perpetual futures position settles gains and losses through the venue rather than by delivering the asset. A trader who posts 4,000 USDT to hold 40,000 USDT of BTC exposure is implicitly using credit: if the price moves far enough against the position, losses exceed the collateral, and someone else (the venue, an insurance fund or profitable traders on the other side) has to absorb the difference.

Liquidation exists to remove the position before that point. The venue sets a maintenance requirement, a cushion of equity that must remain while the position is open, and hands the position to its liquidation engine once equity falls to that level. On Binance, for example, the engine cancels the trader's open orders and tries to offload the position with a single large immediate-or-cancel order, according to the exchange's liquidation protocols.

Equity, initial margin and maintenance margin

The examples in this guide use a linear, USDT-margined BTC perpetual with isolated margin. Position size Q is measured in BTC, prices in USDT per BTC, and margin and P&L in USDT (treated as dollars for readability). P₀ is the entry price and P the current mark price.

• Notional value: Q × P.
• Unrealized P&L: Q × (P − P₀) for a long; Q × (P₀ − P) for a short.
• Equity (margin balance): posted margin + unrealized P&L − fees and funding charged to the position + funding credited to it.
• Initial margin: the collateral required to open the position, here notional at entry divided by leverage L.
• Maintenance margin: the minimum equity the position must keep, here m × Q × P for a fixed maintenance rate m.

The liquidation condition is that equity falls to or below maintenance margin, with both evaluated at the mark price.

Two features of maintenance margin matter. First, it is computed on current notional, so it moves with price: it shrinks as a losing long's notional falls and grows as a losing short's notional rises. Second, real venues tier it. Binance defines maintenance margin as notional times a maintenance-margin rate minus a "maintenance amount," with higher rates for larger notional brackets, and calculates it the same way whatever leverage the trader selects (Binance, Leverage and Margin of USDⓈ-M Futures). Binance expresses the trigger as a margin ratio, maintenance margin divided by margin balance, that reaches 100% (Binance, Liquidation Protocols).

Neither figure is a loss limit. Initial margin is the cost of entry and maintenance margin the line at which the engine acts; the final loss depends on what happens after that line is crossed.

Leverage, notional and the distance to liquidation

Setting equity equal to maintenance margin and solving for price gives the simplified isolated-margin threshold, assuming no fees, funding or added margin:

Long:  P_liq = P₀ × (1 − 1/L) / (1 − m)
Short: P_liq = P₀ × (1 + 1/L) / (1 + m)

The corresponding adverse move is (1/L − m) / (1 − m) for a long and (1/L − m) / (1 + m) for a short. The usable buffer is therefore roughly 1/L minus the maintenance rate, not 1/L. Because the maintenance rate stays fixed while 1/L shrinks, maintenance margin consumes a growing share of the buffer as leverage rises. The table assumes an 80,000 USDT entry and a 0.5% maintenance rate:

At 50x, roughly a quarter of the intuitive "2% cushion" is already spoken for by maintenance margin.

Leverage is a ratio, not a size. Under a single maintenance tier, 10x positions of 0.5 BTC and 5 BTC face the same percentage distance but very different dollar losses, and on a tiered venue the larger one may carry a higher maintenance rate and a closer threshold. What sets an isolated threshold is collateral per unit of exposure: keeping 4,000 USDT of margin while doubling the position to 1 BTC is effectively 20x and moves the long threshold to 76,381.91. In cross margin, the leverage setting does not fix the threshold at all; Hyperliquid's documentation notes that a cross position's actual liquidation price is independent of it, because lower leverage simply uses more collateral (Hyperliquid, Liquidations).

The trigger: maintenance margin measured at the mark price

Major venues test the maintenance condition against a mark price rather than the last traded price. Binance calculates unrealized P&L and liquidation prices from a mark price derived from funding data and a basket of spot-exchange prices, while realized P&L uses the last price (Binance). Bybit states that liquidation is triggered by the mark price and warns that the liquidation price may not appear on its default chart, which plots last-traded prices (Bybit, Liquidation Process). Hyperliquid's mark price combines external exchange prices with its own order-book state.

Two consequences follow. A position can be liquidated even though no trade on the chart ever printed at the threshold, while a brief last-price wick through it may trigger nothing. And a displayed liquidation price is a projection, not a guaranteed trigger level: Binance describes a short with a displayed liquidation price of 17,006 USDT that was liquidated at a mark price of 17,100 USDT after the mark jumped within a second. The engine and the chart may be watching different numbers.

Venues also build reserves and charges into the process. In isolated margin, Bybit adds an estimated fee to close the position to both initial and maintenance margin (Bybit), and Binance deducts a liquidation clearance fee, calculated on notional, from liquidated positions that are not bankrupt.

Example 1: a 10x BTC long

Assumptions (hypothetical): entry 80,000 USDT; long 0.5 BTC; 10x leverage; isolated margin; maintenance rate 0.5% in a single tier with no maintenance amount; taker fee 0.05%.

Step 1: notional and initial margin. Notional = 0.5 × 80,000 = 40,000 USDT. Initial margin = 40,000 ÷ 10 = 4,000 USDT, posted as isolated collateral.

Step 2: equity after an adverse move. At a mark price of 76,000 (−5%), unrealized P&L is 0.5 × (76,000 − 80,000) = −2,000 USDT, so equity is 2,000 USDT. Maintenance margin is 0.005 × 0.5 × 76,000 = 190 USDT. The position survives with 1,810 USDT above the requirement and a margin ratio (maintenance margin ÷ equity) of 9.5%, but a 5% price move has removed half its equity.

Step 3: the simplified threshold. Solve 4,000 + 0.5 × (P − 80,000) = 0.005 × 0.5 × P:

P = (40,000 − 4,000) / (0.5 × 0.995) = 72,361.81 USDT   (−9.55%)

At that price, equity and maintenance margin both equal 180.90 USDT. A widely repeated shortcut, P₀ × (1 − 1/L + m), gives 72,400 because it measures maintenance margin at the entry price instead of at the trigger price. The gap is small here, but the two are different quantities; Bybit's published isolated-margin formula divides by (1 − MMR), as the derivation above does.

Step 4: fees and funding. Assume the 20 USDT opening fee (0.05% × 40,000) is taken from the isolated margin. Assume also that funding is a positive 0.03% per eight-hour interval for ten days, or 30 payments, and that the long pays it. Funding is charged on position value at the mark price (Bybit, Funding Fee Calculation); if the mark stays near 80,000 while the position is held, each payment is about 12 USDT and the total is 360 USDT. Equity falls by 380 USDT with no adverse price move at all:

P = (40,000 − 3,620) / (0.5 × 0.995) = 73,125.63 USDT   (−8.59%)

The threshold moves about 764 USDT closer to the market. Every USDT debited from this position's margin raises its threshold by 1 ÷ (Q × (1 − m)), or about 2.01 USDT. Venues differ on where the debit lands: BitMEX historically took paid funding from an isolated position's margin (BitMEX), while Bybit takes it from available balance first and from the position's margin only when the balance is insufficient. BitMEX ceased trading on September 23, 2026. The constant funding rate is purely illustrative; real rates are reset every interval and can change sign.

Example 2: the comparable short

Assumptions (hypothetical): short 0.5 BTC at 80,000 USDT; 10x; 4,000 USDT isolated margin; maintenance rate 0.5%.

The adverse direction is now up. At a mark price of 84,000 (+5%), unrealized P&L is 0.5 × (80,000 − 84,000) = −2,000 USDT and equity is 2,000 USDT, the mirror image of the long. Maintenance margin, however, is 0.005 × 0.5 × 84,000 = 210 USDT rather than 190, because the short's losing direction increases its notional.

Solving 4,000 + 0.5 × (80,000 − P) = 0.005 × 0.5 × P:

P = (40,000 + 4,000) / (0.5 × 1.005) = 87,562.19 USDT   (+9.45%)

Equity and maintenance margin both equal 218.91 USDT there. The bankruptcy price, where equity reaches zero, is 88,000. The shortcut P₀ × (1 + 1/L − m) would give 87,600.

The short's threshold sits 9.45% from entry, the long's 9.55%. The asymmetry comes from the price dependence of maintenance margin: for a long, the requirement shrinks as price falls, which the (1 − m) term turns into extra room; for a short, it grows as price rises, which the (1 + m) term takes away. The effect is small at a 0.5% rate and larger at moderately higher maintenance rates. A short's loss also has no natural price ceiling.

Costs cut both ways. The 20 USDT opening fee pulls the threshold to 87,522.39. Under the positive funding of Example 1, the short receives 360 USDT, but whether that helps depends on the venue: where received funding is credited to the isolated position, as BitMEX's historical documentation describes, the threshold moves out from 87,522.39 to 88,238.81; where it lands in the available balance, it stays at 87,522.39. If funding were negative at the same magnitude, the short would pay, and with the fee included its threshold would fall to 86,805.97, about 756 USDT closer than the cost-free case.

Example 3: survival depends on the path

Take Example 1's long without fees or funding, so the simplified threshold is 72,361.81. The table follows a hypothetical sequence of mark and last prices at six observation points.

At t3, equity of 100 USDT is below the 180.50 USDT requirement, and the engine takes over. The market then recovers to 81,000. Looking only at entry and final price, the trade would show a 500 USDT profit, but the position no longer exists. If the venue closes it at the bankruptcy price of 72,000, as Bybit does for isolated positions at its lowest risk-limit tier, the realized loss is the full 4,000 USDT, a 4,500 USDT difference from the endpoint view.

The path also shows what a chart can hide. The lowest last price is 72,450, above the threshold, so a last-price chart never shows the line being touched. And the table checks the condition at only six points, while live engines monitor continuously: the breach occurred somewhere between t2 and t3, and a spike that reverses before the next observation can end a position just as decisively. A scenario model built on discrete price paths, as the Perp Mechanics workbench is, can only evaluate the points it is given.

What happens after the trigger

The liquidation price is where forced closing begins, not where losses end. Three further concepts decide the outcome.

The bankruptcy price is the price at which equity reaches zero: 72,000 for Example 1's long, 88,000 for Example 2's short. Between the liquidation price and the bankruptcy price lies the maintenance cushion, 180.90 USDT in Example 1. Partial liquidation reduces a position in steps instead of closing it at once. Insurance funds absorb losses from positions that cannot be closed before equity turns negative, and auto-deleveraging (ADL) closes opposing profitable positions when the fund cannot.

How these fit together is venue-specific. As documented on October 14, 2026:

• Binance sends one large immediate-or-cancel order and stops if the remaining equity, after realized losses and the clearance fee, again covers maintenance. Any unfilled part becomes a bankrupt position that the insurance fund may take over at the bankruptcy price, which can be worse for the trader than the liquidation price. Small positions are more likely to be liquidated in full (Binance).
• Bybit (isolated margin) cancels orders that would enlarge the position. At the lowest risk-limit tier, it closes the position at the bankruptcy price; at higher tiers, it first cuts the position to a lower tier. Better-than-bankruptcy execution sends the remaining margin to the insurance fund; a deficit goes to the fund and then to ADL (Bybit, Liquidation Process; Bybit, Bankruptcy Price).
• Hyperliquid sends market orders to its order book, and collateral left after requirements are met stays with the trader, with no clearance fee. 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. Below two-thirds of maintenance margin, a backstop vault takes over and the maintenance margin is not returned (Hyperliquid).

The same 180.90 USDT cushion can therefore return to the trader, be consumed by a fee or pass to an insurance fund, depending on where the position was held.

ADL is the step that affects traders who were not liquidated. Binance ranks opposing positions by profit percentage multiplied by effective leverage and closes the selected ones at the bankruptcy price of the liquidated order, without a trading fee (Binance, ADL). It is not merely theoretical: an academic preprint found that ADL closed about $2.1 billion of positions on Hyperliquid within 12 minutes on October 10, 2025 (Chitra, arXiv:2512.01112). A profitable, well-collateralized position can be closed because of someone else's failed liquidation.

Losses can also exceed posted collateral. If Example 1's mark price gaps from 74,800 straight to 71,000 and the position is closed there, equity becomes 4,000 + 0.5 × (71,000 − 80,000) = −500 USDT. Who bears that shortfall depends on venue policy and margin mode: Binance describes clearing negative balances through its insurance fund, to the extent possible and subject to account conditions, while in cross margin a deficit first consumes the account's other collateral. An isolated allocation may cap losses as a matter of venue policy, not as a mathematical guarantee.

Why the collateral structure changes the calculation

Every formula above assumes isolated margin, where only the allocated collateral backs the position. In cross margin, Binance's liquidation-price formula draws on the whole cross wallet balance and on the unrealized P&L and maintenance margin of every other contract held; for an isolated position, those other-contract terms are set to zero (Binance, USDⓈ-M liquidation price). Bybit assesses cross-margin risk at the account level and begins its liquidation process, starting with risk-reduction steps, when the account maintenance-margin rate reaches 100%; the displayed liquidation price is therefore not fixed (Bybit, Liquidation Process). A cross-margined position's threshold therefore moves when unrelated positions gain, lose or pay funding. A full comparison of margin modes is beyond this guide's scope; the point here is that no liquidation formula is valid without knowing which collateral stands behind the position.

What a simplified formula leaves out

The thresholds in this guide are mathematical consequences of stated assumptions. A venue's engine adds variables the formula does not capture:

• maintenance rates and maintenance amounts that change with notional, including at the liquidation price itself;
• fee reserves, closing fees and liquidation clearance fees;
• funding timing, sign and the account from which it is debited;
• mark-price construction and sudden mark-price jumps;
• partial liquidation and risk-tier reduction;
• bankruptcy-price takeovers, insurance funds and ADL;
• order-book depth, slippage and price gaps;
• cross, portfolio and coin-margined collateral, where the collateral's own value can move with the price.

The Perp Mechanics workbench models generic isolated-margin scenarios under assumptions the user defines, along discrete price paths. It does not reproduce live exchange liquidation engines, cross-margin portfolios or venue-specific risk tiers, and its outputs should be read as scenario analysis, not as an exchange's liquidation price or an executable exit.

Key takeaways

• Liquidation begins when equity, measured at the mark price, falls to the maintenance requirement, not when the chart's last price touches a number.
• The usable buffer is roughly 1/L minus the maintenance rate, and it differs slightly between longs and shorts because maintenance margin moves with price.
• Fees and funding paid from a position's margin pull its threshold toward the market.
• Survival depends on the whole price path; a favorable final price does not revive a liquidated position.
• Realized losses depend on venue-specific rules for partial liquidation, bankruptcy-price takeovers, fees, insurance funds and ADL, and on execution.