Do liquidation clusters attract price?

Liquidation clusters do not pull the price towards them the way a magnet pulls iron; they mark levels where forced orders would appear if the price got there, and a move that reaches one can speed up for a while. Whether price reaches clusters more often than other levels is an empirical question, and the honest way to answer it is to record every large cluster and count afterwards how many were actually reached.

This guide explains where the "magnet" idea comes from, why it fails as a rule, and how our scoreboard measures it without hindsight.

Where the magnet idea comes from

A liquidation cluster is a price level where an unusually large amount of position size would be liquidated at once. The idea that such levels attract the price rests on three observations that are each true on their own:

• Forced orders add to a move. When the price reaches a long cluster, the liquidated longs are closed with market orders against the bids, which pushes the price further down. A short cluster does the same upwards.¹ A move that touches a dense cluster can therefore run further than it otherwise would.
• Clusters are visible. On Hyperliquid every position is public, so anyone can compute where size would be liquidated. Any trader can place orders around those levels in expectation of the extra flow there.
• Memory is selective. A dramatic move that ran straight through a large cluster is remembered. The many clusters that sat untouched for days and then faded away are not.

Put together, these observations produce the story: price "hunts" liquidations. The first two describe real mechanics. The third is the reason the story feels more reliable than it is.

Why it is not a law

A cluster is a description of current positioning, not a force acting on the price. Several things work against the magnet story:

• Most clusters are never reached. A cluster far from the price needs a large move to be touched. Price is driven by spot flows, news, funding, hedging and much else; a stack of liquidation prices is one input among many.
• Distance does most of the work. A level close to the price is reached more often than a distant one, whatever sits there. Counting hits without the distance next to them says more about distance than about clusters.
• The holders react. The owners of the positions in a cluster see the same levels. As the price approaches, many add margin, reduce size or close, and the cluster shrinks or moves before the price arrives.
• Liquidity is the other half. Even when a cluster is reached, what happens next depends on how much resting liquidity absorbs the forced orders. A deep book can take a large cluster with little effect.

The fading-band effect

On a liquidation heatmap, a cluster is a bright horizontal band. When you watch it over time, three things can happen:

1. The price runs through it. The band disappears from one column to the next, and the liquidation feed shows forced closes at those levels.
2. The band fades without being reached. Its owners added margin, cut size or closed. The band moved away; it was not hit.
3. The band stays and the price never comes. The positions remain open at their levels while the market trades elsewhere.

The second case is the one that most often fools a reader in hindsight. A band that was bright yesterday and is gone today looks like "price took the liquidity" unless you check whether the price ever traded there. How to read the bands themselves, step by step, is in how to read a liquidation heatmap.

How to test it: a hit rate recorded in advance

The fair test of the magnet idea has two rules. Clusters must be recorded before anyone knows the outcome, and every recorded cluster must count, including the ones that were never reached.

That is what our scoreboard does. The method, defined on the cluster hit rate page, works like this:

• At a fixed interval we record the largest liquidation clusters within a fixed band around the mark price, each above a fixed minimum size, together with their distance from the mark. The same bucket is recorded at most once per day.
• A cluster counts as reached when the Hyperliquid mark price touches its bucket: at or below its top edge for a long cluster, at or above its lower edge for a short cluster.
• We check this for three horizons: within 4, 24 and 72 hours. A cluster only counts for a horizon once it is at least that old, so recent clusters cannot inflate or deflate the result.
• The hit rate is the number of reached clusters divided by all recorded clusters, per coin and for all coins together, with the sample size next to it.

The scoreboard checks touches against mark price samples taken at short intervals. A wick between two samples is not seen, so the hit rate is a lower bound. In early weeks with few observations, the rate swings a lot; the sample size tells you how much weight a number can carry.

How to read the result

A hit rate answers a narrow question: of the clusters recorded in the past, how many did the price reach within a given time. It does not answer:

• Why they were reached. A reached cluster may have been reached because of the forced orders around it, or because the market moved for other reasons and the cluster happened to be in the way.
• What happens next time. A past hit rate is a description of past data, not a probability for the next cluster.
• Whether a cluster changes the path. That would need a comparison with similar levels without clusters, at the same distance and in the same volatility.

Read it next to the distance and the volatility of the coin. A high hit rate for clusters that were recorded close to the price in a volatile week says little about clusters far away in a quiet one.

Limits of the data

• Tracked Hyperliquid positions only. Our clusters come from the positions of the Hyperliquid wallets we track. Liquidation levels on centralised exchanges are not public per position, so they are not part of the test.
• Mark price, not last trade. Liquidations on Hyperliquid use the mark price, so the scoreboard uses it too.¹ A trade print beyond a cluster that the mark price never followed does not count as a hit.
• Positions are minutes old. A cluster is recorded from the latest poll of each wallet, so very fresh changes may be missing at the moment of recording.

Liquidation clusters say where forced orders would appear. Whether price gets there is a separate question, and the scoreboard is where we answer it with counts instead of anecdotes.