Leverage and the mechanics of a liquidation cascade
How a modest price move becomes a violent one: margin arithmetic, forced selling into thinning depth, and the loop between them.
A liquidation cascade is a feedback loop in which forced selling pushes the price into the level where more positions must be force-sold. It requires three ingredients that digital-asset markets supply together: widely available leverage, automated liquidation that executes into the open market, and depth that withdraws as volatility rises. None of the three is exotic; the combination is what produces moves that look disproportionate to any news.
The arithmetic that sets the trigger
A leveraged position carries an initial margin requirement and a lower maintenance margin. The distance between the entry price and the price at which equity falls to the maintenance requirement is the position's tolerance, and it shrinks roughly in proportion to leverage. At five times leverage a position is broadly exhausted by a move of about a fifth against it, at twenty times by about a twentieth, before fees and funding. Two account settings change the picture materially. Isolated margin confines the loss to the margin assigned to that position. Cross margin lets the whole account balance support it, which delays the trigger and means a single bad position can consume collateral supporting every other one.
Because most traders choose round leverage multiples and enter near visible reference levels, their trigger prices cluster. Venues can see this clustering in their own books; participants infer it from public liquidation prints. A cluster is not a target in any conspiratorial sense, it is simply a region where a given amount of selling produces more selling than elsewhere, and price tends to travel through such regions quickly.
What the liquidation engine actually does
When the mark price reaches the trigger, the venue takes over the position. Well-designed engines liquidate partially first, closing enough to restore the margin ratio, and they route the resulting order into the same order book everyone else is trading in. That is the essential detail: the liquidation is a market order, indistinguishable from any other aggressive seller, and it consumes market depth. If the position cannot be closed above its bankruptcy price, the shortfall is absorbed by the venue's insurance fund, and if that is depleted, by auto-deleveraging that force-closes profitable positions on the opposite side.
The loop then runs: forced selling consumes the resting bids, the price falls further, the next band of trigger prices is reached, and those positions are liquidated in turn. Each step is a normal, rule-following action by a system doing what it was designed to do.
Why depth disappears at the worst moment
The amplifier is on the other side of the book. A market maker quoting two-sided prices faces inventory risk and adverse selection, both of which rise sharply with realized volatility. The rational response is to widen the spread and reduce size, and in extreme conditions to stop quoting until the picture is clear. Automated risk limits do this without a human decision. So the sequence is not merely that selling meets a fixed book, it is that selling meets a book which is actively shrinking in response to the same volatility the selling created. This is what turns a linear process into a convex one, and it is the reason a cascade liquidation so often ends with a long wick that reverses within minutes: the price traded down to where the last resting bid sat, not to where anyone valued the asset.
The on-chain version has different timing and the same shape
In a lending protocol, borrowing capacity is governed by a collateral factor, and a position becomes eligible for liquidation when its health factor crosses the liquidation threshold. There is no venue engine; independent liquidators compete to repay part of the debt in exchange for collateral at a discount, and that discount is what funds their gas and risk. Three differences change the dynamics.
- The price comes from an oracle, which updates on a schedule or on a deviation threshold rather than continuously. Positions can be underwater before the protocol knows it, and an oracle failure or a manipulated feed can liquidate solvent positions or fail to liquidate insolvent ones.
- Liquidation requires a transaction to land, so network congestion, itself caused by the cascade, is part of the mechanism. During severe events, blockspace becomes the binding constraint and liquidators bid fees against each other, which is one of the clearest expressions of MEV.
- If collateral falls faster than liquidators can act, the protocol is left with bad debt, a shortfall no insurance fund necessarily covers, and one which is socialized across depositors or absorbed by a protocol treasury.
What dampens it, and what does not
Equity markets interrupt this loop with trading halts and limit bands, which force a pause and an auction. Digital-asset venues have no such mechanism; an outage is the only functional equivalent, and it arrives without warning and asymmetrically, since traders who cannot close positions on a halted venue remain exposed elsewhere. What genuinely dampens cascades is duller: lower maximum leverage, position limits scaled to open interest, partial rather than full liquidation, mark prices built from indices robust to single-venue moves, and insurance funds sized against realistic tail moves rather than typical ones. Reading a venue's rules on those five points describes its cascade behavior better than any historical chart.
Reading the aftermath without over-reading it
After a cascade, several artifacts appear in the data. Prints far below the surrounding range on one venue are execution against an empty book, not a price the market agreed on, and any figure computed from a single venue's low, including a drawdown, inherits that artifact. Open interest typically falls sharply, indicating positions closed rather than opinions changed. Volume spikes, which distorts turnover readings and short-window volatility for weeks afterward, since 30-day volatility carries the event until it rolls out of the window. And correlations across assets rise, because forced selling is indiscriminate about fundamentals; a portfolio that looked diversified on ordinary days often was not on the day it mattered.
The incidents pages document specific market dislocations and what caused them, risk collects the per-asset drawdown and volatility measures that these events dominate, and the metric catalog defines exactly which window each figure uses.