Liquidation Engines: Partial, Full and Auction Designs

When margin falls below a maintenance level, the venue closes the position rather than allowing a loss it cannot collect. How it does that — how much it closes, at what price, and who takes the other side — is the difference between an orderly loss and a cascade, and between a bad day for one trader and socialised losses for everyone.

Full liquidation

The simplest design closes the entire position the moment it breaches maintenance margin. It is easy to implement, easy to reason about, and it is the harshest for the trader: a brief wick that touches the threshold closes everything, including the portion that would have been fine a minute later.

It is also the most destabilising at market level, because it puts the maximum possible size onto the market at the worst moment. A cluster of positions liquidating fully in the same price band produces exactly the feedback loop that turns a fall into a cascade.

Partial liquidation

A partial engine closes only enough of the position to restore the required margin ratio. The trader keeps a reduced position and can recover if the price does; the market absorbs less selling per liquidation event.

This is better on both counts and harder to build. The engine must calculate the minimum size that restores health, execute it, and re-evaluate — repeatedly, under load, while prices move. Doing that badly produces repeated partial liquidations that cost more in fees and slippage than a single full close would have.

It also fails differently: if the price keeps falling, partial liquidation buys time rather than solving anything, and a position can be whittled down through several rounds to almost nothing. Whether that is better than one clean close is a matter of preference rather than of design quality.

Auction-based liquidation

Rather than executing directly against the market, the engine offers the collateral to third parties who bid to take it on. The winning bidder repays the debt and keeps the collateral, profiting from the discount.

The advantage is that liquidation does not require immediate market depth — a bidder with capital can absorb the position and unwind it at their own pace. That decouples the venue's solvency from momentary liquidity, which is precisely what fails during a cascade.

The dependency is that bidders must show up. An auction with no participants clears at nothing, and the venue takes the loss. The design's most famous failure mode is exactly this: network congestion raising transaction costs so far that bidding was unattractive, auctions clearing at near zero, and the protocol absorbing bad debt it had assumed the auction would prevent.

The parameter that decides everything

Whatever the design, the liquidation incentive — the discount or bonus the liquidator receives — determines whether anyone acts.

Set too low, liquidators do not participate during congestion, when transaction costs rise and the price may move before their transaction confirms. Positions then fall below the value of their debt and the shortfall lands on the venue or its users. Every large bad-debt event in this category traces to liquidation not happening rather than to liquidation happening.

Set too high, the incentive itself accelerates the damage: more value transferred out of borrowers per event, more collateral onto the market than the debt required, and a faster cascade.

The parameter is a judgement made in advance about which failure the venue fears more, and it is tested only under conditions nobody can simulate honestly.

Setting that parameter is the job of a protocol's risk manager, and the results are observable after the fact: our risk manager ranking scores each firm on the liquidation failures that happened on its watch.

Where the shortfall goes

When liquidation fails to cover the debt, someone absorbs the difference, and venues answer this differently.

An insurance fund built from liquidation penalties absorbs it first, which is the cleanest arrangement while the fund holds.

Socialised loss mechanisms spread the shortfall across profitable traders on the other side — meaning a trader who was right can be paid less than they earned because someone else was wrong and the engine could not close them in time.

Protocol reserves and, ultimately, token issuance are the on-chain equivalents. All of these are worth knowing about before they are used, and all of them are published.

• Which design does the venue use, and does it close positions fully or partially.
• What is the liquidation penalty, and is it large enough to attract liquidators when transaction costs spike.
• Where does a shortfall land — an insurance fund, socialised losses, or protocol reserves.
• How large is the insurance fund relative to open interest, which is public on most venues.
• What happened during the last severe drawdown. The record exists and is the only real test.