Why a liquidation is not a stop-loss

On a chart, a forced sale looks like any other sale: a print at a price, a volume bar slightly taller than its neighbours. The difference is not in what reaches the tape. It is in who owns the order, and that difference decides how the next minute goes.

A stop-loss is a decision with three parts, and all three belong to the person who placed it: the level that triggers it, the size it releases, and the freedom to cancel it at any moment up to the trigger. Somebody who does not like how the book looks can move it, shrink it, or take it off entirely and wait.

A liquidation has none of the three. The trigger sits in the venue's margin formula, the size is whatever the position happens to be, and there is no cancel. Binance states the condition as collateral — initial collateral plus realised and unrealised profit and loss — falling below the maintenance margin, and evaluates it against the mark price rather than the last traded price, so a thin print on one venue does not by itself set it off.

Twenty times leverage is not a five percent buffer. A position opened at twenty times puts up one twentieth of its notional value, so the intuition is that a five percent move against it wipes the stake out. But liquidation does not wait for the stake to reach zero; it happens when equity falls below the maintenance margin. On Hyperliquid that margin is half the initial margin at the asset's maximum leverage — 1.25 % for an asset that offers forty times, 16.7 % for one that offers three. Against a forty-times asset, a twenty-times position starts with 5 % and is closed at 1.25 %, so the move that ends it is 3.75 %, not 5 %.

The distance shrinks faster than the leverage grows. Run the same arithmetic across the ladder on that asset: ten times leaves 10 % less 1.25 %, or 8.75 % of room; twenty times leaves 3.75 %; forty times leaves 1.25 %. Doubling leverage from twenty to forty does not halve the distance to liquidation, it cuts it to a third. This is why the interesting positions cluster in the last stretch before the threshold, and why a move that would be unremarkable on spot is terminal here.

The threshold is not a constant of the trader. Venues raise the margin requirement as a position grows: Binance adjusts risk and leverage by absolute exposure, so the larger the position, the higher the required margin and the lower the leverage permitted. The same trader at the same nominal leverage therefore sits at a different distance from liquidation depending on size, which means the largest positions — the ones whose forced sale moves price most — are also held to the strictest thresholds.

Once the threshold is crossed, the venue sends the order. Hyperliquid describes it as market orders sent to the book for the position; Binance issues an immediate-or-cancel order to offload it. Either way it is a market order: it names no price, and it takes whatever depth is there.

Which is the problem, because of when it arrives. Resting depth on a large venue mostly belongs to market makers, and a market maker's rational response to one-directional flow is to widen or step back — the subject of the article on depth against pool liquidity. Forced orders are the most one-directional flow there is. So the order that cannot be delayed reaches the book at the moment the book is least able to absorb it.

From there the loop is arithmetic rather than intent. The fill consumes levels, the price it leaves behind is lower than the price that triggered it, and that lower price is the liquidation threshold of some other position held at a higher entry or a higher multiple. Nobody coordinates this and nobody has to. Each forced sale is simply an input to the condition that produces the next one.

Venues know it and blunt the edge. Binance tries first to reduce the margin deficit with one order rather than closing the whole position outright, and stops there if that suffices. Hyperliquid closes only twenty percent of a position over 100,000 USDC at first, then waits thirty seconds before the remainder is exposed. Both make the individual order smaller. Neither changes the mechanism.

When a position is closed below the price at which its collateral would have covered it, somebody absorbs the shortfall. On Binance that is the futures insurance fund, which bears the loss arising from the bankrupt position. This is the part of the machinery that works quietly and is invisible from a chart.

When the fund is not enough, the venue reaches for auto-deleveraging: the matching engine closes bankrupt positions and, alongside them, opposing positions belonging to traders who are not in trouble at all. That is the point where forced closure stops being a consequence of one trader's leverage and becomes something that happens to a counterparty who did nothing wrong. It is disclosed, it is rare, and it is worth knowing about before it becomes the explanation for a position that vanished.

Hyperliquid has an equivalent backstop: if equity falls below two thirds of the maintenance margin and the book has not absorbed the position, it transfers to the liquidator vault, and the maintenance margin is not returned to the trader. The published rules of both venues are worth reading once in calm conditions, rather than being discovered in the fifteen minutes when they apply.

The cascade simulator here runs exactly this loop and nothing more: you state a book, a set of position buckets with their entries and leverage, a maintenance margin rate and an opening shock, and it walks the passes until no further position crosses its threshold. It observes no venue, reads no open interest and predicts nothing. It answers what the arithmetic does to a scenario you supplied.

Three real effects are deliberately absent from it, and all three make a genuine cascade smaller than the simulated one: insurance funds, auto-deleveraging, and the resting orders that appear precisely because the price moved. A number from that page is an upper bound on a stated scenario, not a forecast of a market.