Liquidations and cascades

What is margin, exactly?

Margin is collateral posted against a position, not a payment for it. When you open a leveraged position, the venue reserves an initial margin: the minimum equity required to carry that notional. Leverage is simply the ratio between the two — a $50,000 position against $5,000 of equity is 10×, whatever the interface calls it.

Your equity moves with the position. Unrealised profit adds to it, unrealised loss subtracts from it, and funding payments and fees are deducted as they occur. The single number the risk engine watches is the relationship between that equity and what it currently requires you to hold — and requirements are computed against notional, which moves too.

Isolated and cross margin change what is at stake. Isolated margin ring-fences a fixed amount per position: the loss is capped at that amount and the position is closed when it is exhausted. Cross margin lets the whole account balance back every position, which means a position survives longer — and that a single bad position can consume collateral that was implicitly supporting the others. Neither is safer in the abstract. Isolated caps the loss per position; cross reduces the chance of being closed out by noise, and raises the chance that everything goes at once.

What is maintenance margin, and when exactly are you liquidated?

Initial margin is what you need to open. Maintenance margin is what you need to stay open, and it is smaller — typically a percentage of notional that rises in tiers as position size increases, because a larger position is harder for the venue to unwind. When your equity falls below the maintenance requirement, the position is handed to the liquidation engine. There is no grace period and no discretion.

One detail that saves accounts: the trigger is the mark price, not the last traded price. The mark is derived from an index across several spot venues, sometimes with a smoothing of the basis. This exists specifically so that a violent print on one venue — a fat finger, a thin book at three in the morning — does not liquidate positions everywhere. It also means the price you see printed on your own venue’s chart may be below your liquidation level without you being liquidated.

The arithmetic is worth doing once by hand, because the intuition it corrects is a common one: that N× leverage means a 100/N percent move wipes you out. It does not — the maintenance requirement gets there first.

Run the same calculation across leverage settings, with the same illustrative 0.5% maintenance rate, and the shape of the risk becomes obvious: the distance to liquidation collapses far faster than leverage rises. The liquidation price calculator runs it for any entry, leverage and maintenance rate.

참고 These are illustrative figures computed from the example above, not measurements. At 50×, ordinary intraday noise is enough to close the position — which is the whole reason clusters of liquidations exist within a percent or two of the current price.

What actually happens when a position is liquidated?

The risk engine takes over. Your open orders on that symbol are cancelled, and the position is closed — on many venues in stages, with partial liquidations reducing size until the account is compliant, rather than all at once. The closing order is aggressive by necessity: the venue cannot afford to rest a limit order and hope.

If the position closes better than the bankruptcy price, the remainder flows to an insurance fund. If it closes worse — which is what happens in a fast market with no depth — the insurance fund covers the shortfall so the winning counterparty is still paid. This is the part most traders never think about until it matters: the fund is finite.

When it is not sufficient, venues fall back on auto-deleveraging: profitable positions on the opposite side are closed, by force, at the bankruptcy price, usually ranked by profit and leverage. You can be right, be up, and have your position taken off you because somebody else on the other side blew up and the fund could not absorb it. It is rare and it is not hypothetical, and it belongs in any honest description of the risk of holding leveraged exposure through a violent move.

A liquidation also costs more than the loss itself. Venues charge a liquidation or clearance fee on top, which is why the realised loss on a liquidated position is generally worse than the same position closed voluntarily one tick earlier.

How does one liquidation trigger the next?

Because a liquidation is a market order, and market orders move price. Follow the loop one step at a time.

1. 01Price moves enough to breach the maintenance threshold of the most leveraged positions on one side.
2. 02The risk engine sends aggressive orders in the same direction as the move — selling into a falling market, buying into a rising one. This flow is price-insensitive: it must be done, at whatever the book offers.
3. 03That flow consumes depth. The book was already thinner than usual, because market makers widen when volatility jumps and adverse selection spikes.
4. 04The resulting price move breaches the next band of maintenance thresholds, which sit just below in a market where leverage is distributed continuously.
5. 05Repeat, at machine speed, until the flow exhausts the positions in range — or until liquidity arrives because the price has become attractive enough for someone to want it.

Three things make the loop worse than the arithmetic suggests. Cross-margin accounts spill across symbols: a loss in one contract can force liquidation in an unrelated one held by the same account, which is how a move in one asset propagates to others with no fundamental connection. Stop-loss orders trigger on the same levels, adding discretionary forced flow to the mechanical kind. And the mark-price mechanism, which protects against single-venue wicks, does not protect against index-wide moves — when every constituent falls together, the mark falls with them.

The loop is also self-limiting, which is why cascades end abruptly. Forced flow is finite: once the positions within range are closed, the flow stops entirely, and whatever depth arrived to absorb it is now sitting on one side of a price that overshot. That asymmetry is the reason the reversal is often as violent as the move.

Where does the fuel come from in the first place? Crowded leveraged positioning, which is exactly what funding rates measure.

How do you read a liquidation heatmap?

Start with what it is not. No venue publishes the liquidation price of individual accounts. A liquidation map is therefore an estimate, built by taking observable quantities — open interest, position and trade data, price history — assuming a distribution of leverage over the positions opened at each price, and projecting where the maintenance thresholds of those hypothetical positions would sit. Brightness is estimated notional that would be forced out at that price.

That makes it an estimate built on assumptions, and the assumptions are the whole product. Two maps of the same market can disagree because they assume different leverage buckets, different margin modes, or different treatment of positions that have since been closed. A map that does not tell you its assumptions is decoration.

Read properly, it answers one question well: if price reaches here, roughly how much price-insensitive flow would be generated? That is genuinely useful, because it tells you where liquidity is guaranteed to exist and where slippage is likely to be violent. It is a map of magnets and trapdoors, not a forecast.

• Dense clusters below in an uptrend are fuel for a flush, not a prediction of one.
• The removal of a cluster is information. Once a band has been swept, the fuel there is spent, and the same level behaves completely differently on the second visit.
• Symmetry matters. Clusters on both sides at similar distance describe a market that can move violently in either direction; clusters stacked on one side describe an asymmetric market.
• Distance matters more than size. A large cluster 15% away is irrelevant to today; a modest one 1.5% away is today’s entire story.

참고 Verify the map against what actually happened: after a move through a cluster, did open interest fall and did the liquidation feed print? If a “cluster” is repeatedly traded through with no observable effect, the map’s leverage assumptions do not match that market.

Why do wicks reach into liquidation clusters?

Two mechanisms, and they reinforce each other. Neither requires anyone to be manipulating anything, though the second certainly permits it.

The mechanical reason

A cluster is, by construction, a price at which a large quantity of forced market orders will be generated. When price arrives there, that flow fires into a book that has just been widened by the volatility of getting there. Price-insensitive orders meeting reduced depth produce a large print in a short time — the wick. Then the flow stops, because the positions are gone, and the price snaps back to where willing participants are actually trading. The wick is long because the flow was inelastic, and it reverses because the flow was finite.

The intentional reason

The clusters are estimated from public data, so everyone with a terminal can see roughly the same map. A participant who wants to buy size has a strong reason to prefer buying it from a forced seller: a liquidation is the most reliable counterparty in the market, because it must trade regardless of price. Pushing a market the last fraction of a percent into a cluster is not free, but the payoff is a large fill at a price nobody would otherwise offer. On venues without a surveillance regime, assume this is a normal part of the landscape.

Both mechanisms explain why clusters coincide with round numbers, prior highs and lows, and obvious technical levels: that is where stops are placed, where leveraged entries were taken, and therefore where the thresholds accumulate. The level is not magic. It is where the forced flow lives.

And it explains the standard misreading. “The market rejected that level” attributes intent to a print that was, mechanically, a forced seller meeting an empty book and then stopping. The candle is a description of a liquidity event, not a vote.

How do you read a liquidation feed without fooling yourself?

Public liquidation streams are among the most quoted and least understood data in crypto. Several caveats change the interpretation materially:

• Public feeds are throttled and coalesced on some venues, so the published total is a lower bound on what was actually liquidated — not a census. Treat cross-venue totals as an index, not an amount.
• One printed event is one order, which may be one slice of a partial liquidation of one account. Counting events is not counting accounts.
• The printed price is a venue-specific convention — it may be the order price, the bankruptcy price, or the fill price. Comparing across venues without checking is comparing different quantities.
• The direction label is the position being closed, not the order’s side. A “long liquidation” is a forced sell.

The honest confirmation is open interest. A genuine flush removes positions, so open interest falls; a large liquidation print with flat open interest means the positions were replaced as fast as they were closed, which is a very different market. Pair the feed with open interest and with the book before drawing a conclusion, and treat any single venue’s number as a sample.

What actually changes your exposure to this?

Mechanically — and this is a description of how the machinery responds, not a recommendation about what to do with your money:

• Leverage sets the distance to your threshold, and the relationship is convex: each step up shortens the distance by more than the last, as the table above shows.
• Maintenance requirements rise in tiers with position size. A position twice as large is not twice as risky; it can require a proportionally larger buffer.
• Adding margin and reducing size both improve the ratio, and they are not equivalent. Adding margin keeps the same exposure with more capital at risk; reducing size lowers the exposure itself.
• Fees and funding erode the buffer continuously. A position held for a week at high leverage is closer to its threshold than it was on day one even if price has not moved.
• Auto-deleveraging exists. Even a correctly margined winning position can be closed for you if the insurance fund is exhausted on the other side.

None of this is advice, and EdgeMarket does not give any: it is the mechanism, stated plainly, so that the number on your screen means what you think it means.

자주 묻는 질문

Why was I liquidated when the price never touched my liquidation level?

Almost always because the trigger is the mark price rather than the last trade on your venue, and the two differ during fast moves. Fees and funding also shift the level over time, so a liquidation price quoted when you opened the position is not the level in force a week later.

Are liquidation heatmaps accurate?

They are estimates. Venues do not publish account-level liquidation prices, so every map infers them from assumed leverage distributions. Use them for the shape — where forced flow would be dense — and not as a ledger of real orders.

Do cascades happen in prediction markets too?

Not in the same way. A spot prediction-market share is fully collateralised, so there is no margin engine to force a close — the worst case is that the price falls and nobody is bidding. The mechanics are covered in how prediction markets work.