EVA Terminal · LearningРусский

Section 12 of 18

Liquidations: mechanics and maths

Why the exchange closes a position itself, what maintenance margin is, how to estimate the liquidation price and why it is closer than “100% / leverage”, how a stop differs from a liquidation, and what follows the heaviest hours.

What a liquidation is

A leveraged position is worth more than you put in: $1,000 of margin at 10× holds a $10,000 position. The loss is computed on the whole position but covered from the margin. A 1% move against you is −$100, i.e. 10% of margin. Loss in percent of margin = price move × leverage.

The exchange cannot let a loss exceed the margin: someone else — the winning side — would have to cover the gap. So every position has a line at which the exchange closes it itself, without asking the owner. That is a liquidation — a forced close of a position when almost no margin is left. The line is set not by zero margin but by maintenance margin — the minimum balance that must remain on the position while it is open (more in the second lesson).

Most large exchanges check the trigger not against the last trade but against a fair (mark) price — a price the exchange derives from several venues. That way one random trade in a thin book does not wipe out everyone's positions.

A position can only be closed by an opposite trade. A long is closed by selling, a short by buying. Therefore:

Long liquidations happen when the price falls and add sells exactly where the price is already falling. Short liquidations happen on rises and add buys. These are market orders nobody chose: the owner might have preferred to keep holding, but the decision was made for them. On the tape they look like ordinary aggressive trades, and the exchanges' separate liquidation feed shows which ones were forced. How such closes chain into each other and what a map of possible levels shows is the topic of section 13.

Isolated margin — the position has its own margin, and only that is lost in a liquidation. Cross margin — the whole free account balance supports the position: liquidation comes later, but the whole account is at risk, not one position.

Where liquidations are in EVA:

Coverage differs by exchange: Bybit, Gate and Kraken Futures send every liquidation; OKX — a history that is included while the sample is fresh; Binance and Bitget — at most one liquidation per second per coin, so some small ones are not visible there. Hyperliquid's public data does not mark liquidations, so it is not in the totals. Under the total it says how many coins and exchanges it covers — totals from different services can only be compared at the same coverage.

Maintenance margin and the liquidation price

A position has two margins. Initial — what you must post to open: position value / leverage. Maintenance — the minimum the balance must not fall below: position value × the maintenance rate. The exchange sets the rate, usually fractions of a percent for small positions, and it rises in steps with position value: a large position needs a bigger buffer because closing it at market is harder.

Liquidation comes when the margin balance (posted margin plus unrealised result) falls to the maintenance level. For a long with entry E, leverage L and maintenance rate m this gives E/L + (P − E) = m × P, hence:

long: P ≈ E × (1 − 1/L) / (1 − m);
short: P ≈ E × (1 + 1/L) / (1 + m).

For a mental estimate a simplification is enough: distance to liquidation ≈ 1/L − m. A 10× long at m = 0.5%: 10% − 0.5% = 9.5% below entry.

Illustrative: a 10× long from $100,000, isolated margin, 0.5% maintenance90,00095,000100,000entrystop: −5%liquidation ≈ 90,452“100% / leverage” = 90,0009.55% to liquidation, not 10%: part of the margin is maintenance, which the exchange won't let you lose.A stop closes the position on your plan; liquidation — when almost no margin is left.
A 10× long from $100,000: stop at −5%, liquidation ≈ $90,452 (−9.55%), not $90,000 as “100% / leverage” promises.

Computing the liquidation price

Illustrative example, isolated margin, maintenance rate 0.5%.

BTC long, entry $100,000, 10× leverage. Initial margin is 10% of the position. P = 100,000 × 0.9 / 0.995 ≈ $90,452. That is 9.55% below entry. Simplified: 10% − 0.5% = 9.5%, i.e. $90,500 — a $48 difference, irrelevant for an estimate.

BTC short, entry $100,000, 10× leverage. P = 100,000 × 1.1 / 1.005 ≈ $109,453, 9.45% above entry.

ETH short, entry $3,000, 20× leverage. P = 3,000 × 1.05 / 1.005 ≈ $3,134, 4.48% above entry. Margin is $150 per ETH, yet the price only needs to travel $134 to liquidation.

The exchange computes more precisely than this formula: it uses its own maintenance tier, reserves the fee for the forced close and compares against the mark price. So the real liquidation price is usually a little closer to entry still. The exact number is always shown by the exchange in the position card; the formula is for understanding where it comes from and sanity-checking it.

Why is liquidation closer than “100% / leverage”? Because “100% / leverage” is the move at which the margin would be gone entirely. But the exchange does not wait for zero: it closes the position when the balance reaches maintenance, to have time to close at market without going negative. On top of that, the closing fee is reserved. Both eat into the buffer.

At low leverage this part is invisible, at high leverage it is decisive. At m = 0.5%: at 5× liquidation is 19.6% away instead of 20% (98% of the buffer remains), at 20× — 4.52% instead of 5% (90%), at 50× — 1.51% instead of 2% (75%), at 100× — 0.50% instead of 1%: maintenance margin eats half the buffer. And as position value grows, m rises and the buffer shrinks further.

Maintenance tiers are worth checking at least once on your exchange. Illustrative example: up to $50,000 of position value the rate is 0.5%, while at a tier of several million it is 1% or more. A 20× long on a small amount is liquidated about 5% − 0.5% = 4.5% from entry. The same 20× long at the 1% tier — already at 5% − 1% = 4%. Nothing changed but the amount, yet the buffer shrank by a ninth. That is why exchanges cap leverage for large positions: at a 2.5% rate, 40× would leave 2.5% − 2.5% = 0 to liquidation.

How much of “100% / leverage” remains to liquidation (0.5% maintenance)5×20.00% → 19.60%98%10×10.00% → 9.55%95%20×5.00% → 4.52%90%50×2.00% → 1.51%75%100×1.00% → 0.50%50%Grey bar — “100% / leverage”, coloured — the real room to liquidation. Illustrative.
The share of the “100% / leverage” distance that actually remains to liquidation at 0.5% maintenance.

Margin, leverage and the stop

Three things move the liquidation price.

Cross margin pushes liquidation further away using the whole free balance. It feels safer, but the cost of a mistake is different: with isolated margin you know in advance how much you can lose on this position, with cross you do not.

A stop and a liquidation are different things. You place the stop order, at a level you chose, with a loss you computed in advance. The exchange triggers liquidation when almost no margin is left, and you lose practically all of the position's margin; what remains of maintenance is usually not returned to you but goes to the close.

Illustrative example. A 10× long from $100,000, margin $10,000. Stop at $95,000 (−5%): loss $5,000 — half the margin, plus entry and exit fees (at 0.05% per side — about $50 for the entry and as much for the exit). After the stop about $4,900 is left in the account, and what to do next is your decision. Liquidation at ≈ $90,452: the entire $10,000 margin is gone.

A stop is not perfect either: it becomes a market order on trigger and in a fast market fills with slippage. But a stop placed well before the liquidation price is a boundary you set. A position whose only protection is liquidation means the exchange decides how big the loss is.

Loss in percent of margin = price move × leverage. 3% against a 20× long is 60% of margin, even though “the price only moved three percent”.

What follows a heavy hour

BTC's heaviest liquidation hours

Kontur archive, 4 September — 12 October 2026. We summed BTC liquidations by hour (the archive records closes from $100,000) and took the 5% of hours with the largest total — 35 hours, from ≈ $6.54M per hour and up. For each, we compared the price direction in that hour and in the next.

19 of 34 is about 56%. “After mass liquidations there is always a bounce” is false: continuations were even more frequent. “If it is wiping people out, it will keep doing so” is not a rule either: 15 of 34 reversed. A heavy hour says many leveraged positions were force-closed and the leverage skew on that side shrank. After that — it varies, and 34 cases are too few to conclude anything from 56% versus 44%. The figure shows the first case in time of each outcome, not the most striking ones.

BTC: the heaviest liquidation hours (top 5%). The first case of each outcome in time.next hour continued · 19 of 34next hour reversed · 15 of 34Kontur archive 04.09–12.10.2026 · bars — liquidations per 5 minutes · dashed — start of the heavy hour

Source: kontur archive ~/eva-kontur/archive, of_2026-09-04 … of_2026-10-12 (39 files), liq events for BTCUSDT, sum of u per hour; top 5% of hours; price at the start of the hour, 1 and 2 h later (± 10 min). Script dev/kurs_statistika.py with ARH=~/eva-kontur/archive → zamery/kurs/statistika-kontur.json

How to read liquidations without turning them into a signal:

Common mistakes

The section test: 12 questions
Tricky questions and a review of every answer. The test, your progress and the section order are available after signing in: a section opens after passing the previous one, the pass mark is 80%. We teach you to read the market, not to guess: there are no signals or forecasts here.
Take the test — log in

The whole course

  1. How the market works: spot, futures, leverage — What exactly you buy on spot and on futures, how to calculate profit and loss with leverage — as a percentage of margin and of the position, who makers, takers and aggressors are, and why price moves only through trades.
  2. Orders and execution: price, fees, slippage — Limit, market and stop orders, take-profit; how a market order fills across several prices and how to compute the average fill price; maker and taker fees on entry and exit — charged on the position, not the margin; slippage and common calculation mistakes.
  3. The order book: bids, asks, spread and depth — How the order book works: levels, best prices, the spread and the mid price; depth and how a market order eats through levels; thin and thick books in the EVA archive; multi-exchange books and what the book does not show.
  4. Liquidity walls: what rests and what gets pulled — What a dense order and a wall are, how a filled wall differs from a pulled one, what spoofing is, how to read book imbalance by distance bands and the heatmap — and why a wall is an intention, not money.
  5. The tape: orders, merging, pace — What the tape and prints are, why an exchange sends one order as a dozen trades, how EVA merges them back, what “large” means on its own venue, how to read tape pace — and what the tape does not show.
  6. Volume and the volume profile — What bar volume says and does not say, why volumes of different exchanges do not match, how the volume profile, POC and value area (VAH/VAL) work, how to read high volume without movement, and session profiles.
  7. Delta and CVD: who was pressing — Bar delta is market buys minus market sells; CVD is delta added up. Why “there were more buyers” is the wrong phrase, how to read the slope, why futures and spot are shown separately, and what delta does not show.
  8. Divergences, absorption, CVD by trade size — When price and CVD go different ways, when market flow is met by limit orders, and which orders are pressing — small, medium or large. With honest statistics from our archive: 53 continuations against 54 reversals.
  9. Footprint: inside the candle — What happened between the open and the close: market sells and buys at each price level, the candle's POC, the diagonal imbalance and stacked imbalances — with calculations, a real BTC bar from the archive and an honest count over a week.
  10. Open interest — How many futures positions are open, how every trade changes that number, why OI in dollars misleads when the price moves, and what our archive honestly says about OI growth.
  11. Funding — Why a perpetual future has a payment between longs and shorts, who pays whom, how to compare rates of exchanges with different intervals, what it costs a position, and what our archive says about “overheated” funding.
  12. Liquidations: mechanics and maths — Why the exchange closes a position itself, what maintenance margin is, how to estimate the liquidation price and why it is closer than “100% / leverage”, how a stop differs from a liquidation, and what follows the heaviest hours.
  13. Cascades and the liquidation map — How one forced trade pushes the price into the next, why the chain breaks, what the estimated liquidation map really draws and why a liquidation total always names its exchanges. With 378 bursts from our archive.
  14. Exchange flows and whales — What a coin transfer to or from an exchange shows through our node, why a transfer is a chance to sell rather than a sale, how to read stablecoins, ETFs and exchange reserves with their delays, and who the off-exchange “whales”, US government wallets and Hyperliquid whales are. With two days, 12–13.10.2026, worked through.
  15. Macro and news — The release calendar: time, forecast, previous value and the actual from the primary source; why the market reacts to the deviation from expectations rather than the number itself; what happens to the book before a release; how to read dominance and the TOTAL caps; news and retellings of retellings. With 19 US releases from EVA's calendar worked through.
  16. Risk: position size and stop — How much you may lose on one trade, where the stop goes and how these two numbers give the position size; what leverage really does, how much of R fees eat, and why the percentages on the exchange screen are not percentages of the account.
  17. Risk: streaks, expectancy, journal — How much a system makes per trade on average, why losing streaks are inevitable, how recovery from a drawdown is computed, what to record in a trading journal and which mistakes break even profitable trading.
  18. Using EVA as a whole — The tab order and the day's order — from general to particular; the “?” sign and the “what it does not show” point; how to read “no data”, the source and the time of data; narrow alerts; replay for an honest review of an entry, and a journal in R.