How much of a position does a haircut liquidation engine actually close?

Only as much as the math requires. In a $100,000 two-position account with $60,000 long BTC (down 5%) and $40,000 long ETH (down 2%), account equity fell to $2,866.67 against a $3,000 maintenance requirement — a 0.956x health ratio. Restoring that to a 1.02x target required shedding $189.54 of maintenance requirement, which meant closing $6,318.08 of the BTC position: 10.53% of it. The ETH position wasn't touched at all.

Can a haircut liquidation close more than the losing position?

Yes, once the losses are severe enough that fully closing every losing position still isn't enough. In a modeled $150,000 account where a $100,000 BTC long lost $8,000 and equity fell to $500 against a $3,750 requirement (a 0.133x health ratio), the engine had to close 100% of the BTC position and then reach into the ETH position — which was profitable by $1,000 — and close 61.9% of that too. Total notional closed: $130,952.38, which is 130.95% of the original $100,000 BTC position alone, or roughly 87% of the full $150,000 account.

Why does a profitable position ever get partially closed?

Because the maintenance-margin math doesn't care about the sign of a position's PnL, only its notional and maintenance rate. Closing $1 of any position's notional frees up maintenance-requirement room, whether that position is up or down. Once every losing position in the account has been fully liquidated and a shortfall still remains, the only notional left to trim belongs to positions that were, until that moment, winning trades.

What target health ratio does a haircut engine restore an account to?

There's no universal number — it's a buffer above the bare 1.0x minimum so the account doesn't immediately re-trigger on the next price tick. The two scenarios here used 1.02x and 1.05x targets, both plausible ranges. A tighter target trims less notional per liquidation event but risks re-triggering sooner; a looser target trims more up front and buys more room before the next check.

Is a haircut liquidation always cheaper than a full liquidation?

Only in the range where it's actually mild. The 10.53% case is a clear win over a full-position close. The 130.95%-of-notional case is not obviously better than a full liquidation of the original position — it just happens to distribute the damage across two positions instead of concentrating it in one, and it still leaves 38.1% of the ETH position and $0 of the BTC position standing. If equity had reached zero or below, no amount of trimming would have worked at all — that's outright insolvency, not a haircut case.

I Ran the Same Losing Portfolio Through Two Liquidation Engines. One Closed 10.5%. One Closed 130.95%.

Published 17 September 2026

Every perp exchange I've built a liquidation calculator for so far shares one design: breach maintenance margin on a position, and that position closes. All of it, usually with a penalty on top. BULK's perpetual DEX, which leaned hard into "portfolio-wide risk management" as its pitch this year, does something different — it computes the smallest amount of notional it needs to trim, anywhere in the account, to get equity back above the maintenance requirement, and stops there. I built a calculator around that mechanism and ran two portfolios through it to see exactly how much smaller "smallest amount" actually is, and where it stops being small.

Case 1: the mild breach — 10.53% closed, 89.47% survives

Account: a $60,000 long BTC position (15x leverage, $4,000 posted margin) down 5% from a $60,000 entry to $57,000, and a $40,000 long ETH position (also 15x, $2,666.67 posted margin) down 2% from $3,000 to $2,940. Both carry a 3% maintenance rate. Total collateral posted: $6,666.67. Net unrealized loss: -$3,800.00 ($3,000 on BTC, $800 on ETH). Account equity: $2,866.67. Total maintenance requirement: $1,800 (BTC) + $1,200 (ETH) = $3,000.00. Health ratio: 2,866.67 ÷ 3,000 = 0.9556x — below the 1.0x minimum, liquidation triggers.

A full-liquidation exchange closes the entire $60,000 BTC position here (it's the one that breached), or in some designs the whole $100,000 account. The haircut engine instead targets a 1.02x restored ratio: M_target = 2,866.67 ÷ 1.02 = $2,810.46, so it needs to shed 3,000 − 2,810.46 = $189.54 of maintenance requirement. Since BTC's maintenance rate is 3%, that's 189.54 ÷ 0.03 = $6,318.08 of BTC notional — 10.53% of the position. The other 89.47% of BTC stays open. The ETH position, the smaller loser, isn't touched at all.

Case 2: the severe breach — the "winning" position gets cut too

Same shape, worse numbers. A $100,000 long BTC position (20x, $5,000 margin) drops 8% from $60,000 to $55,200, losing $8,000. A $50,000 long ETH position (20x, $2,500 margin) actually gains 2%, up $1,000. Collateral: $7,500. Net PnL: -$7,000. Equity: $500.00. Maintenance requirement: $2,500 (BTC) + $1,250 (ETH) = $3,750.00. Health ratio: 0.1333x — a severe breach, not a marginal one.

Targeting a 1.05x restored ratio needs M_target = 500 ÷ 1.05 = $476.19, a shortfall of 3,750 − 476.19 = $3,273.81 to shed. BTC's full maintenance-requirement capacity is only $2,500 (its entire position), so the engine closes 100% of the $100,000 BTC position — takes the full $2,500 — and still has $773.81 of shortfall left. It moves to ETH next, the only capacity remaining, and closes 773.81 ÷ 0.025 = $30,952.38 of it: 61.90% of a position that was profitable when the liquidation started. Total notional closed across the account: $130,952.38 — 130.95% of the original BTC position by itself, or about 87% of the original $150,000 combined account.

Scenario	Health ratio	BTC closed	ETH closed	Total notional closed
Case 1 — mild	0.9556x	10.53% ($6,318.08)	0%	$6,318.08
Case 2 — severe	0.1333x	100% ($100,000)	61.90% ($30,952.38)	$130,952.38

The part that isn't intuitive: the loser gets closed first, always

The engine I modeled works worst-loss-first — it fully exhausts the maintenance-requirement capacity of the biggest loser before touching anything else. That's why ETH survives untouched in Case 1 despite also being a loser: BTC alone had enough capacity ($1,800 of requirement, more than the $189.54 needed) to cover the shortfall. It's also why ETH gets cut in Case 2 even though it was profitable at the moment of liquidation — BTC's entire $2,500 of capacity wasn't enough, and closing notional reduces the maintenance requirement by notional × maintenance_rate regardless of whether that notional belongs to a winning or losing trade. A profitable position in the same account is not protected from a severe-enough shortfall elsewhere in the portfolio.

The one case where none of this math helps: equity at or below zero before any trimming starts. Both scenarios above had positive equity ($2,866.67 and $500.00) — the haircut mechanism only works in the range where there's still something to restore. Below zero, it's outright insolvency, and the whole account goes, exactly like a full-liquidation exchange. Run your own two-position account through the same worst-loss-first math on the portfolio margin partial liquidation calculator — it shows the same waterfall table these two cases came from, with your own numbers.

Trade where the calculators point

Trade on Bybit — lower fees, free to join → Binance → OKX → KuCoin → 📈 TradingView 🔒 NordVPN