The Trapdoor Nobody Sees Until It Opens
You've borrowed against your crypto, the position looks comfortable, the buffer feels wide, and then one bad afternoon later you're watching a bot drain your collateral while the price keeps falling. That's not bad luck. That's a cascade, and it has a very specific anatomy.
Cascading liquidations happen when one wave of forced collateral sales pushes prices down far enough to trigger another wave of liquidations, which pushes prices down further still, and so on until either the selling exhausts itself or the market finds a genuine buyer willing to absorb the volume. The price doesn't just drop. It falls in steps, each step kicking a new trapdoor open beneath a new set of borrowers.
Understanding why requires looking at exactly how lending protocols decide when to pull the trigger.
The Math That Starts Everything
Every major DeFi lending protocol (Aave, Compound, MakerDAO) keeps a running ratio for each loan: the value of what you've borrowed divided by the value of the collateral backing it. This is the loan-to-value ratio, or LTV. When you open a position, the protocol sets a liquidation threshold, a specific LTV percentage above which your position becomes eligible for liquidation.
Aave has historically set liquidation thresholds in the 75-85% range depending on the asset. Cross that line and external liquidators, bots almost always, can repay part of your debt in exchange for a chunk of your collateral at a discount, typically a 5-10% liquidation bonus. That discount is the profit motive. Without it, no one would do the liquidating.
Here's a worked example. Suppose you deposit 10 ETH as collateral when ETH is worth $2,000 each. Your collateral is worth $20,000. You borrow $14,000 in stablecoins, an LTV of 70%. The liquidation threshold is 80%, meaning your collateral would need to fall to $17,500 (14,000 / 0.80) before you're at risk. That's an ETH price of $1,750. Feels safe.
Now ETH drops to $1,750. Your LTV hits 80%. A liquidator bot repays $7,000 of your debt and takes $7,700 worth of ETH (the extra $700 is the 10% bonus). It sells that ETH immediately on a DEX to capture the profit. That sale pushes the ETH price down a fraction. Small on its own. But you're not the only borrower with a threshold near $1,750.
Why One Liquidation Begets the Next
Think of the order book for a thinly traded asset as a staircase with uneven steps. Every liquidation is someone taking a sledgehammer to one step. Fine, if the steps are far apart. Catastrophic, if they're stacked.
The clustering problem is the real villain here. Borrowers don't set their liquidation thresholds randomly. They cluster at round numbers and at the maximum borrowing capacity the protocol allows, and when an asset has been trading in a tight range for weeks, a large proportion of those positions will have been opened near the same price. Their liquidation thresholds land within a narrow band just below the current market price. It's limescale in a kettle: invisible while conditions are stable, suddenly everywhere when the heat comes on.
Protocols like MakerDAO publish their liquidation wall data publicly. Analysts tracking these walls will sometimes identify a band, say $50 wide on a $1,500 asset, where $200 million in collateral would become eligible for liquidation if price crosses it. When price crosses it, $200 million in collateral hits the market in minutes. That selling pressure drops the price into the next band.
The speed matters enormously. Ethereum blocks process roughly every 12 seconds. A sophisticated liquidation bot submits its transaction the instant the price oracle updates, often in the same block. Human borrowers scrambling to add collateral are almost always slower. The bots are faster by design, and that gap is not a coincidence.
The Oracle Problem Hiding Inside the Cascade
Here's the wrinkle most explainers skip entirely. The price that triggers a liquidation isn't the real-time spot price on a centralized exchange. It's the price reported by the protocol's oracle, often a time-weighted average price (TWAP) or an aggregated feed from something like Chainlink. That design choice is intentional: a TWAP is harder to manipulate with a single large trade.
The catch: it creates a lag. During a fast crash, the oracle price trails the spot price by anywhere from a few seconds to a few minutes depending on the protocol's configuration. This lag can actually compress liquidations. When the oracle finally catches up after a sharp drop, it may cross multiple liquidation thresholds simultaneously, triggering a burst of liquidations all at once rather than a gradual sequence. One catch-up tick. Ten thousand positions eligible. Every liquidator bot on the network racing for the same collateral.
That race itself is a problem. When many bots compete to liquidate the same position, they bid up gas fees, sometimes to extraordinary levels. During severe cascades, gas fees on Ethereum have historically spiked by multiples within a single hour. Smaller liquidators get priced out; only the most capital-efficient bots survive. The collateral still gets sold. Just by fewer hands, faster.
What People Get Wrong About This
The popular narrative is that cascading liquidations are caused by panic selling. Incomplete. Panic selling among spot holders contributes to the initial price drop, sure. But the cascade itself is mechanical, not emotional. It would happen even if every human stopped trading. The bots would carry it.
People also assume that being well-collateralized means being safe. Consider two borrowers: Maria opens a position at 50% LTV with a liquidation threshold at 80%. She has enormous buffer. Ravi opens at 75% LTV, threshold at 80%, on the same asset. Ravi is one bad week from the threshold. But here's what's underappreciated: if a cascade starts and the price falls 40% in an hour, Maria might also get liquidated, not because her individual position was reckless, but because the liquidation of Ravi and thousands like him moved the price past her threshold too. The cascade doesn't care that she was conservative. It just needs the price to reach her number.
Are you sure you know where your liquidation threshold actually sits right now? If your current LTV is below 50% on a volatile asset, you have meaningful breathing room. Above 70%, you're playing closer to the edge than most people realize.
The Self-Reinforcing Engine
Cascading liquidations are, at their core, a feedback loop with an automatic trigger. Price falls. Thresholds get crossed. Collateral gets sold. Price falls further. The loop only breaks when one of three things happens: the price drop stops attracting new liquidatable positions (the cascade has eaten through the vulnerable layer), a large external buyer absorbs the selling pressure, or the protocol itself intervenes through some circuit-breaker mechanism, which most current lending protocols simply don't have.
Some newer protocols are experimenting with gradual liquidation systems, liquidating small percentages of a position continuously rather than all at once when a threshold is crossed. The theory is that smaller, distributed sales create less downward pressure than one massive liquidation event. Whether that holds under genuine stress conditions is still an open question, and frankly, optimism about untested mechanisms is how people get surprised.
The cascade is a feature of the design, not a bug that can be patched with a single line of code. It exists because the incentive to liquidate quickly is baked into the liquidation bonus. Remove the urgency and the bonus, and liquidators won't show up at all. Keep it, and you get the trapdoor. Every lending protocol is living with that tradeoff, and none of them has solved it yet.