What Is Liquidation in DeFi?

In brief

• An initial loan-to-value limit controls new borrowing; a distinct liquidation threshold controls when an existing position can be forcibly closed.
• Aave V3 expresses account safety as a health factor. Below 1, the documented V3 position is eligible for liquidation.
• A close factor limits how much debt a liquidator may repay in one event. It can be partial, full or state-dependent and is not universal.
• A liquidation bonus or discount compensates liquidation activity and market risk. It is an economic cost to the borrower, but protocol fees can mean that borrower penalty and liquidator revenue are not identical.
• If collateral sale proceeds do not cover debt and costs, the system has bad debt. Reserves, safety modules, auctions, governance or suppliers may ultimately bear it, depending on the design.
• Aave V3 direct repayment-and-collateral transfer, Compound III account absorption, and Sky Dutch auctions are materially different liquidation systems.

From borrowing limit to liquidation boundary

Protocols usually build a buffer between the amount an account may initially borrow and the point where it can be liquidated.

Loan-to-value (LTV) sets borrowing power. If collateral is worth $10,000 and its borrowing LTV is 70%, the simplified maximum new borrow is $7,000.

Liquidation threshold sets the safety boundary for existing debt. If the threshold is 80%, that simplified position becomes liquidatable when debt exceeds $8,000 relative to the protocol’s collateral valuation.

The gap is intentional. Interest can increase debt, collateral can fall, a borrowed asset can rise, and transactions take time to include.

For an Aave V3 account with several collateral assets, the documented health factor is:

[ \text{health factor} = \frac{\text{total collateral value} \times \text{weighted average liquidation threshold}} {\text{total borrow value}} ]

Suppose the threshold-weighted collateral value is $8,000 and debt grows to $8,100. The health factor is about 0.988, below 1, so the account is eligible under the V3 rule.

The protocol does not know a universal “true” dollar value. It applies configured parameters to oracle prices. That is why a mathematically correct health factor can still be economically unsafe.

What can push a position across the line

• Collateral falls. The standard ETH-collateral/stablecoin-debt example.
• Debt asset rises. Borrowing a volatile asset creates risk on both sides of the ratio.
• Interest accrues. Even flat prices can gradually increase debt enough to lower health.
• A stablecoin depegs. “Stable against stable” can move sharply if collateral weakens or debt strengthens.
• Oracle data updates. A delayed feed can move the protocol’s recorded value abruptly when it catches up.
• The account changes. More borrowing, collateral withdrawal or disabling an asset as collateral reduces the cushion.
• Parameters change. Governance may alter risk settings, although protocols can design separate borrowing and liquidation factors to avoid immediately liquidating existing accounts.

Crossing the line makes liquidation valid. It does not guarantee that a liquidator will arrive, that the transaction will execute at once, or that enough value will be recovered.

Close factor, bonus and penalty

Three terms are often collapsed into “the liquidation fee.”

Close factor is the maximum fraction of an account’s debt that can be repaid or closed in one liquidation call. A partial close can restore some safety and avoid unnecessary sale; a larger close can be necessary when an account is deeply unhealthy or too small to leave cleanly. Aave V3’s current documentation describes state-dependent partial or full liquidation logic. Exact cutoffs and dust rules are version-sensitive and should be read from the relevant deployment.

Liquidation bonus is extra collateral made available relative to the debt repaid. It gives external actors an incentive to pay gas, source repayment assets, accept price movement and compete for inclusion.

Liquidation penalty is the borrower-side economic haircut. In some designs a portion can go to the protocol, so it need not equal the liquidator’s net profit. The liquidator must still account for swap fees, price impact, network fees, failed attempts and hedging.

These parameters trade off competing goals. A tiny incentive may fail during volatility; a large one makes liquidation costly and can encourage aggressive competition. A large close factor can protect solvency quickly but sell more collateral than needed to restore a buffer.

Aave V3: direct liquidation of part or all of a position

Aave V3 uses an account health factor and per-reserve liquidation thresholds. Once health falls below 1, an external liquidator can repay eligible debt and receive corresponding collateral at a discount determined by the liquidation bonus and protocol rules.

The borrower’s debt and collateral both decrease. If only part of the debt is closed, the remaining account may become healthy or may remain eligible for another liquidation. Current Aave documentation includes conditions under which a greater share of debt can be liquidated; those thresholds should not be copied as timeless constants.

This is the model many beginner explanations describe, but it is not the model for every protocol.

Compound III: absorb first, sell collateral later

Compound III separates initial borrow collateral factors from higher liquidation collateral factors. Failing the initial borrow check can stop an account from increasing debt without making it immediately liquidatable. Liquidation eligibility uses the separate liquidation factors.

When eligible, any address can call absorb. The protocol:

• transfers the underwater account’s debt to the protocol;
• uses base-asset reserves to repay that borrow;
• takes ownership of the account’s collateral;
• leaves the user with no remaining debt and, typically, a residual base-asset balance after the documented penalty.

The caller that triggers absorption does not simply repay debt and receive collateral in the Aave pattern. Compound records liquidator points and gas usage. The protocol later makes seized collateral available for purchase at a configured discount when reserves are below their target.

This two-stage process moves immediate balance-sheet risk to protocol reserves and separates account absorption from discounted collateral sale. Calling every Compound III collateral buyer “the liquidator who repaid the account” is inaccurate.

Sky: transfer to the protocol and run an auction

Sky’s vault liquidation documentation describes another model. When a vault is insufficiently collateralized, its collateral is transferred and its debt is assigned to the protocol. The Dog liquidation contract starts a Dutch auction through the collateral-sale system, seeking stablecoin proceeds to cancel the debt and liquidation penalty.

A Dutch auction begins at a configured price and decreases according to its rules until a buyer takes collateral or the auction needs to be reset. Auction participants may use on-chain liquidity, including atomic funding, but proceeds and timing depend on market depth.

Sky limits how much debt can be active in liquidation globally and per collateral type through Hole-style parameters. These caps are intended to avoid overwhelming external liquidity. They also mean an unsafe vault may not be immediately sent to auction if the relevant capacity is already full.

Sky vault auctions, Compound III absorption and Aave direct collateral transfer all aim to control undercollateralized debt, but they allocate timing, inventory and execution risk differently.

When liquidation leaves bad debt

Bad debt is debt the protocol can no longer fully match with recoverable collateral value under its rules. It can arise when:

• collateral gaps down faster than oracle updates and transactions can react;
• a collateral token depegs, freezes, becomes unsellable or loses most of its value;
• liquidator incentives are too small for gas, price impact and market risk;
• auctions or collateral sales have too little demand;
• chain congestion, sequencer failure or contract pauses delay execution;
• a manipulated or faulty oracle admits too much borrowing or triggers the wrong action;
• protocol code or accounting fails.

Overcollateralization reduces the probability of loss; it does not prove that collateral can be sold for the oracle value.

Who absorbs a shortfall depends on design. A protocol may use accumulated reserves, a surplus buffer, a safety module, governance-controlled recapitalization, debt auctions or a reduction in what suppliers can redeem. Backstops are finite mechanisms, not insurance guarantees.

Liquidation as MEV

A profitable liquidation is ordering-sensitive. Searchers monitor oracle updates and account health, simulate calls and compete to be included before rivals. The available bonus or discounted collateral can become MEV distributed among searchers, builders, validators or leaders through fees and bids.

Flash loans can let a liquidator source repayment capital, receive collateral, sell it and repay within one atomic transaction. That makes liquidation possible without pre-funding the whole debt, but it does not make the trade risk-free: the entire path must satisfy contract conditions and cover fees or revert.

Compound III again differs because the absorber does not receive seized collateral in the same direct exchange. Its ordering incentive and later collateral-purchase economics are not identical to Aave’s.

Cascades and market depth

Liquidations often require collateral to be sold. Large or clustered sales can move DEX prices. If lending oracles observe those markets, lower prices can weaken other accounts and trigger additional liquidations.

BIS and OECD research documents this feedback channel in observed DeFi markets. It is a risk pattern, not a law:

• liquidators may hedge or retain collateral instead of immediately selling;
• deeper markets may absorb sales with limited movement;
• oracle design may damp short-lived price changes;
• caps, auctions and circuit breakers may spread execution over time.

The same mitigations can create delay and basis risk. There is no parameter set that guarantees fast liquidation, minimal borrower loss and zero market impact in every crash.

Oracle correctness is only one requirement

A liquidation system needs more than an honest price:

1. Correctness: the price reflects the intended market and denomination.
2. Freshness: the value updates quickly enough for the collateral’s volatility.
3. Market depth: collateral can actually be sold near that value.
4. Incentives: someone benefits enough to execute under stress.
5. Chain availability: transactions can land before the shortfall widens.
6. Backstop capacity: residual loss has a defined, funded path.

A stale-high price delays liquidation and can create bad debt. A stale-low or erroneous price can liquidate an otherwise healthy position. A correct midpoint from a thin market may still overstate realizable sale value.

What borrowers can and cannot control

Before liquidation eligibility, repaying debt or adding eligible collateral generally improves the safety ratio. Reducing volatile debt exposure or maintaining a wider buffer can reduce sensitivity to price moves and interest.

Those actions are not guaranteed rescues:

• the oracle may update before the transaction lands;
• gas or priority fees may spike;
• a sequencer, RPC or front end may be unavailable;
• deposits, repayments or withdrawals may be paused;
• another liquidation may execute first;
• adding more correlated collateral can increase money at risk without removing the underlying exposure.

Automated alerts observe data; they do not reserve blockspace or stop protocol execution. Liquidation has no grace call from a collections department unless a particular protocol explicitly implements one.

Liquidation is also not impermanent loss. IL compares AMM liquidity with holding. Liquidation is forced collateral disposition under a credit rule.