Lending without credit assessment requires collateral, and the whole design follows from how that collateral is managed when prices move.
Overcollateralisation
Borrowers deposit more value than they borrow.
Which substitutes for credit assessment, since there is no identity and no recourse.
The ratio required depends on the volatility of the collateral, with more volatile assets requiring larger buffers.
Health factor
A single number expressing how close a position is to liquidation.
Which combines collateral value, borrowed value and the asset-specific thresholds.
It changes continuously as prices move, which is why positions can be liquidated while the borrower is asleep.
The liquidation mechanism
When the threshold is crossed, anyone can repay part of the debt and claim collateral at a discount.
Which is the incentive that makes liquidation happen without a central operator.
The discount, generally a defined percentage, is the liquidator's profit and the borrower's loss.
Why liquidation bots exist
The opportunity is competitive and time-sensitive.
Which means automated systems monitor positions continuously and compete to execute first.
This competition is generally beneficial for the protocol, since it ensures liquidations actually occur.
Bad debt
Where collateral value falls below debt before liquidation completes, the protocol carries a loss.
Which happens during rapid price movements, during network congestion, or where collateral is illiquid.
Reserve funds and insurance mechanisms exist to absorb this, with varying capacity.
Interest rate models
Rates generally adjust with utilisation, rising as available liquidity falls.
Which incentivises repayment and new deposits when the pool is nearly fully lent.
Many models include a kink where rates rise sharply above a target utilisation.
The withdrawal problem
Depositors can only withdraw what is not currently lent.
Which means high utilisation can temporarily prevent withdrawal.
This is the same maturity mismatch that conventional banking has, implemented transparently.
Isolated and cross collateral
Whether all deposits back all borrowing, or specific assets are ring-fenced.
Which determines whether a failure in one asset can affect the whole protocol.
Isolated designs limit contagion and reduce capital efficiency, and both approaches are in use.
This describes mechanism and is not advice about borrowing or lending.
Close factor
How much of a position can be liquidated in one transaction, generally capped at a proportion.
Which limits the damage from a single liquidation and can require multiple rounds to restore health.
Some designs liquidate fully below a second threshold, which produces a sharper outcome for the borrower.
Collateral quality
Assets differ in liquidity, volatility and correlation with what is borrowed.
Which is why parameters are set per asset rather than uniformly.
Illiquid collateral can be liquidated at a price the market cannot actually absorb, which produces bad debt even when the position was nominally healthy.
Recursive positions
Depositing, borrowing, and redepositing to amplify exposure.
Which multiplies both returns and liquidation sensitivity substantially.
These positions unwind violently in a falling market, and the cascading effect has been observed in several downturns.
Governance risk
Parameters including thresholds and supported assets are generally set by governance.
Which means the rules governing your position can change while it is open.
Proposals are published in advance, and few borrowers follow them.
Managing a position
Monitoring health, maintaining a buffer well above the threshold, and understanding that price feeds update discretely rather than continuously.
Which is the practical difference between a position that survives volatility and one that does not.
Oracle dependency
Liquidation decisions depend entirely on the price feed the protocol uses.
Which makes feed quality a direct determinant of whether positions are liquidated correctly.
Feeds that update on deviation thresholds can lag rapid moves, producing liquidations at prices that no longer reflect the market.
Network congestion
Liquidations require transactions, and transactions require block space.
Which means the mechanism degrades precisely during the volatility that triggers it.
Documented events show liquidations failing to execute during congestion, leaving protocols with undercollateralised positions.
Fixed and variable rates
Most protocols use variable rates that change with utilisation.
Which means borrowing cost is not predictable over time.
Fixed-rate designs exist and require matching lenders and borrowers on term, which is harder to bootstrap.
Flash loan interaction
Uncollateralised same-transaction borrowing is offered by several lending protocols.
Which has legitimate uses in refinancing and arbitrage.
It also removes capital as a barrier to attacks on other protocols, which is where most of its notoriety comes from.
Comparing protocols
Parameters per asset, oracle configuration, reserve size, historical bad debt and governance arrangements.
All of which are public and vary considerably between protocols offering the same nominal service.
Practical position management
Keeping a substantial buffer, avoiding correlated collateral and debt, and setting alerts on health factor.
Which prevents most liquidations that were avoidable.
This describes how these systems operate and is not advice about using them.
Historical events
Documented episodes of cascading liquidation show the mechanism amplifying price moves.
Which is a systemic property rather than a flaw in any one protocol.
Understanding it explains why leverage concentration is watched as a market indicator.
Summary
The rules are explicit, public and enforced automatically, which is the design's main virtue and the source of its sharp edges.