What is liquidation in crypto? Health factors & more

Liquidation is the moment crypto’s leverage machinery takes your collateral, and it happens two very different ways: exchanges force-closing leveraged trades, and DeFi lending protocols auctioning borrowers’ collateral to keeper bots. This guide explains both systems, the health factor math, the bonus liquidators earn, why liquidations cascade into crashes, and how to read the daily liquidation numbers everyone quotes and few understand.

On a single day this week, roughly $410 million of leveraged crypto positions were liquidated inside 24 hours, most of them longs, and the number scrolled past in headlines the way weather does. Days above a billion dollars are not rare; across 2025, more than $150 billion in positions were liquidated across venues. Liquidation is the most routine catastrophe in crypto, the mechanism by which every form of on-chain and exchange leverage enforces its one non-negotiable rule: the debt gets paid, and if you will not pay it, your collateral will.

What the headlines flatten is that liquidation in crypto is actually two distinct systems wearing one name. The first lives on derivatives exchanges, where leveraged perpetual-futures positions are force-closed when losses approach the trader’s margin. The second lives in DeFi lending protocols like Aave and Compound, where overcollateralized loans are enforced by an open market of bots, called keepers or liquidators, that repay underwater borrowers’ debts in exchange for their collateral at a discount. The two systems share a purpose, keeping lenders and venues solvent without trusting anyone, and differ in almost every mechanical detail, and understanding both is close to understanding how crypto’s entire credit machine holds together.

This guide covers the whole territory: why liquidation exists at all, the derivatives version in brief with its margin math and mark prices, the DeFi lending version in depth with health factors, thresholds, bonuses, and the keeper economy, the anatomy of a liquidation cascade, what the insurance funds and bad-debt backstops do when liquidation itself fails, and the practical playbook for keeping your own positions alive.

Why liquidation exists: solvency without trust

Every liquidation system answers the same question: how does a lender who cannot sue you, cannot call you, and does not know who you are make sure a loan gets repaid? Traditional finance answers with identity, courts, and credit scores. Crypto answers with overcollateralization and automation: you post more value than you borrow, and the moment the cushion between your collateral’s value and your debt shrinks toward zero, the system sells your collateral before the cushion is gone. Done correctly, the lender never takes a loss, because the sale happens while the collateral still covers the debt.

Everything else is implementation detail, and the details matter enormously. Liquidate too late and the protocol eats bad debt; liquidate too early and borrowers are punished for noise; misprice the collateral for one block and either error happens at scale. Liquidation design is where a credit system’s real risk decisions live, which is why it deserves more attention than the afterthought paragraph it usually gets.

Liquidation on exchanges: the derivatives version

On derivatives venues, liquidation is the endgame of leverage. A trader posts margin and opens a position several times that size; the exchange continuously marks the position against a manipulation-resistant mark price; and when losses erode the margin to the venue’s maintenance threshold, the engine seizes and closes the position. At 10x leverage a roughly 10 percent adverse move is fatal; at 50x, about 2 percent. The full mechanics, initial versus maintenance margin, mark versus index price, cross versus isolated margin, are covered in this publication’s perps guide, and two points from that machinery matter for what follows.

First, the mark price, an index-anchored, smoothed price, exists so that a momentary wick on one venue’s order book cannot liquidate everyone; you are liquidated against the market’s consensus price, not the last print. Second, when a position is so far underwater that closing it at market recovers less than the debt, exchanges reach for backstops: an insurance fund absorbs the shortfall, and if the fund is exhausted, auto-deleveraging forcibly closes profitable traders on the opposite side to balance the books, a mechanism that cost profitable traders over $50 million during one violent stretch in late 2025. Exchange liquidation, in other words, is a private matter between you and the venue’s risk engine, with socialized losses as the final resort.

Liquidation in DeFi lending: the health factor and the keepers

DeFi lending liquidation is a different animal, public, permissionless, and run by an open market of hunters, and it is the version most explanations skip.

Start with the loan. On a protocol like Aave, you deposit collateral, say ETH, and borrow against it, say USDC, up to a loan-to-value cap well below 100 percent. Each collateral asset carries a liquidation threshold, the LTV at which the position becomes seizable; for major assets this might sit around 80-83 percent, meaning a loan is safe while the debt stays below that fraction of the collateral’s value. The protocol compresses your entire position into one number, the health factor: the value of your collateral weighted by its liquidation thresholds, divided by your debt. Above 1, you are safe. At exactly 1, your position crosses the line. Below 1, anyone on earth may liquidate you.

And anyone does, because liquidation is a paid job. A liquidator repays some or all of your debt to the protocol and receives, in exchange, your collateral worth what they repaid plus a liquidation bonus, typically around 5 percent for major assets and more for volatile ones. Repay $10,000 of an unhealthy borrower’s USDC debt, receive roughly $10,500 of their ETH; the borrower’s remaining collateral shrinks by the bonus, which is the penalty they pay for crossing the line. Most protocols cap how much of a position can be liquidated in one bite, commonly 50 percent of the debt, called the close factor, so a borrower who dips just below 1 is partially liquidated back to health rather than wiped out, though deeply underwater positions can be fully seized.

The hunters are keeper bots: automated programs that watch every loan on every protocol, simulate health factors against live prices, and race to submit liquidation transactions the instant a position crosses 1. The race is ferocious, the bonus goes to whoever lands first, gas auctions and the private relays of the MEV supply chain decide winners by milliseconds, and the capital to repay the debt is very often flash-borrowed, so the entire operation, borrow the repayment, liquidate, sell the seized collateral, repay the flash loan, pocket the bonus, completes inside one atomic transaction. This is the part outsiders find alien: DeFi does not employ a risk department. It posts a bounty and lets mercenaries keep the system solvent, and it works, most of the time, better than the systems it replaced.

One number rules everything above: the price. Health factors are computed from oracle prices, so the entire lending-liquidation apparatus inherits the oracle’s integrity. A stale or manipulated feed liquidates healthy borrowers or spares doomed ones, and oracle failure is behind a large share of DeFi’s historical bad-debt events, which is why serious protocols use aggregated, median-filtered feeds and why borrowers should know which oracle guards their loan.

A worked example: one loan’s journey to liquidation

Numbers make the machinery concrete, so follow a single position from opening to seizure.

A borrower deposits 10 ETH as collateral with ETH at $1,800, collateral value $18,000, on a protocol where ETH carries an 82.5 percent liquidation threshold. They borrow 10,000 USDC, a 55.6 percent loan-to-value, comfortable territory. Their health factor at opening is the threshold-weighted collateral over debt: 18,000 times 0.825, divided by 10,000, equals 1.485. The position can absorb a meaningful drawdown; solving for the price at which the health factor hits 1 gives the liquidation price: debt divided by threshold divided by ETH quantity, 10,000 / 0.825 / 10, which is about $1,212. ETH must fall roughly 33 percent from entry before the keepers come.

Now the market delivers exactly that. ETH slides over two weeks to $1,250, health factor 1.03, and the borrower, watching, does nothing, reasoning the bounce is near. A weekend wick takes ETH to $1,195 for eleven minutes. At $1,212 the health factor crossed 1, and within a block or two a keeper acts: with a 50 percent close factor, it repays 5,000 USDC of the debt and, with a 5 percent bonus, claims $5,250 worth of ETH, about 4.39 ETH at the wick price. The borrower now holds 5.61 ETH backing 5,000 USDC of debt; the health factor resets to roughly 1.11, alive but poorer. The eleven-minute wick cost them $250 in bonus plus the spread on collateral sold at the local bottom, and if the fall had continued, subsequent liquidations could each take their bite until nothing remained.

The counterfactuals are the lesson. Repaying 2,000 USDC of debt at any point before the wick would have lifted the liquidation price to about $970, far below the wick; adding 2 ETH of collateral would have done similar work; and either action would have cost transaction fees measured in dollars against a penalty measured in hundreds. Liquidation almost never happens without a long, visible approach, the health factor decays in public, on-chain, for anyone to see, and the borrowers it takes are overwhelmingly the ones who watched it come.

The same arithmetic scaled up explains the professional side. A keeper repaying $5,000 for $5,250 earned 5 percent on capital deployed for one block, capital that was itself flash-borrowed, meaning the return on the keeper’s own funds, gas and infrastructure aside, approaches the absurd. That yield is the bounty that guarantees no unhealthy loan survives long, and competition for it is why the bounty has not needed to be larger.

Cascades: how liquidations become crashes

Liquidations do not just respond to price moves; past a threshold, they cause them, and the feedback loop is the mechanism behind many of crypto’s sharpest candles.

The anatomy is simple. A price drop pushes a tranche of leveraged positions past their liquidation points. Liquidation is executed by selling the collateral or closing the longs, which is sell pressure, which pushes the price lower, which liquidates the next tranche, which sells, and so on down the order book. Thin liquidity amplifies every leg, because each forced sale moves the price further, the slippage cost that large orders always pay becoming, in aggregate, the crash itself. The cascade ends where the leverage does: when the liquidatable positions are exhausted, the forced selling stops, and price frequently snaps back, leaving a wick that marks exactly how deep the leverage ran. Funding rates, open interest, and liquidation heatmaps let traders estimate where those clusters sit, which is why the derivatives data services publish liquidation maps and why sophisticated actors sometimes push price toward known clusters to set the dominoes off.

DeFi lending adds its own cascade variant with correlated collateral. When a widely used collateral asset depegs or gaps, every loan built on it sickens simultaneously; the 2022 stETH episode, where a liquid staking token’s discount stressed a leverage loop built on it, remains the canonical case study of one asset’s wobble propagating through lending markets as a synchronized health-factor collapse. The lesson generalizes: your liquidation risk is not just your own leverage but everyone else’s leverage in the same collateral.

When liquidation fails: bad debt and backstops

The system’s last chapter is what happens when selling the collateral does not cover the debt, because prices gapped too fast or liquidity vanished. On exchanges, the insurance fund pays, then auto-deleveraging conscripts the winners. In DeFi, the shortfall becomes bad debt on the protocol’s books, and each protocol has its own waterfall: reserve funds accumulated from fees, safety modules of staked tokens that can be slashed to cover deficits, or, historically and controversially, governance deciding who eats the loss. A protocol’s bad-debt record and backstop design are, alongside its oracle, the two lines of due diligence that matter more than its advertised yields, because they are the difference between a lender that survived its worst day and one that socialized it.

One structural nuance completes the lending picture: not all collateral is liquidated the same way. Fixed-bonus seizure of the kind in the worked example is the dominant design, but several protocols instead auction the collateral, Dutch auctions that start above market and decay until a keeper bites, which returns more value to borrowers in calm conditions and can struggle in chaos, as an infamous episode of zero-bid auctions during a 2020 crash proved when network congestion let liquidators win collateral for nothing. Auction versus fixed-bonus, close factors, per-asset thresholds, and oracle choice together form each protocol’s liquidation personality, and experienced borrowers read those parameters the way credit analysts read covenants, because they are the covenants.

Reading the liquidation tape like a professional

The daily liquidation statistics are among crypto’s most quoted and least understood numbers, and extracting their real information takes three habits.

First, read the ratio before the total. A $400 million day that is 63 percent longs says the market fell into a crowded long book; the same total at 85 percent shorts, like the recent session where Bitcoin short liquidations dominated on a squeeze higher, says the opposite: bears were crowded and the move ran them over. The skew identifies which side was overextended, which is the tradable information; the headline total mostly measures volatility times leverage.

Second, treat totals as minimums. Public figures aggregate what venues report, and reporting conventions differ, some exchanges publish only samples of liquidation events, so true forced-closure volume typically exceeds the printed number. Comparisons across time are still meaningful because the undercounting is roughly consistent; comparisons across venues are not.

Third, connect the tape to positioning data. Liquidations are the discharge; open interest and funding rates are the stored charge. Rising open interest with extreme funding is leverage accumulating on one side, the precondition for a cascade; a liquidation spike that coincides with an open-interest collapse means the leverage actually left the system, which is what durable local bottoms and tops are made of, whereas a spike that barely dents open interest means the crowd reloaded and the fuel remains. The heatmap services that estimate where liquidation clusters sit at each price complete the picture, showing the magnets that sharp moves gravitate toward.

None of this predicts direction on its own; all of it describes the terrain, and traders who read the terrain stop being surprised by which moves extend and which reverse violently at a wick.

Keeping your positions alive: the practical playbook

For a borrower or leveraged trader, all of the machinery above compresses into a few habits. Know your number: the liquidation price on a perp, the health factor on a loan, and the oracle both are computed from. Size for the gap, not the trend, because liquidation happens at the wick, not the close, and weekend and low-liquidity hours produce the worst wicks. Prefer isolated margin when experimenting, so one dead trade cannot drain an account, and keep a repayment buffer ready, since topping up collateral or repaying a slice of debt is dramatically cheaper than the liquidation bonus. Watch the crowd as well as yourself: extreme funding, ballooning open interest, and dense liquidation clusters near price are the weather report for cascades. And read the daily liquidation totals correctly: $410 million liquidated, 63 percent longs is not a death toll but a positioning report, telling you which side was crowded, how much leverage just left the system, and, often, why the price wicked exactly where it did.

Liquidation is easy to resent and hard to replace. It is the reason DeFi lending survived drawdowns that killed centralized lenders whose loan books ran on trust and phone calls, and the reason a perp exchange can offer 50x leverage to anonymous traders and remain solvent by Tuesday. The machine is impartial to the point of cruelty, it will take a sleeping borrower’s collateral over a five-minute wick, and its impartiality is precisely the property everything else is built on. The practical wisdom is old and short: the machine cannot be negotiated with, so stay out of its reach.

A closing word on the system’s deeper logic. Liquidation is crypto’s replacement for the entire apparatus of credit assessment, and the trade it makes is time for capital. A bank spends weeks deciding whether you will repay over years; a protocol spends no time at all deciding, demands surplus collateral instead, and enforces continuously. The design is capital-inefficient by construction, you must lock more than you borrow, and in exchange it achieves something credit systems never had: solvency that does not depend on being right about anyone. Every innovation in the space, cross-margin, isolated pools, dynamic thresholds, liquidation auctions that replace fixed bonuses with competitive bidding to return more value to borrowers, is an attempt to soften the capital inefficiency without surrendering the trustlessness, and the frontier of lending design is exactly that negotiation. Understanding liquidation is therefore not just self-defense for the leveraged; it is understanding the load-bearing wall of the whole on-chain credit system, the mechanism every yield, every stablecoin loan, and every leveraged position in DeFi quietly rests on. The wall holds because the machine is merciless, and the machine is merciless so that no one has to be trusted, which is, for better and worse, the entire proposition this industry was built to test.

And for readers who came to this piece from a headline, the translation service one last time: liquidations hit $X billion is not news that money vanished, most of it moved from the margin accounts of the crowded side to the other side of their trades, and it is not a prediction of anything. It is an after-action report on where leverage lived, published by the only market on earth candid enough to print its casualties in real time.

Disclaimer: This article is for educational purposes only and does not constitute investment advice. Leverage and DeFi lending carry significant risk, including total loss of collateral. Protocol parameters cited are typical values as of July 8, 2026, and vary by platform. Always do your own research.

Frequently asked questions