Liquidation data vs liquidation heatmaps

Two kinds of data with different roots

Liquidation totals and liquidation heatmaps both use the word liquidation, but they come from different sources. Totals count forced closures that exchanges reported after they happened. A heatmap draws price levels where forced closures could happen, based on a model.

The first is an observation. The second is an estimate. An observation can be incomplete, but it describes events that took place. An estimate can look precise, but it depends on assumptions that nobody can check directly.

Keep the two apart in your notes. Write reported liquidations when you describe totals. Write modelled levels when you describe a heatmap. Mixing the words leads to mixing the claims.

Totals describe what was reported; a heatmap describes what a model assumes.

What a reported liquidation is

A leveraged position uses collateral called margin. The exchange requires a minimum level of margin, called maintenance margin. When losses push the margin below this level, the exchange closes the position by force. This forced closure is a liquidation.

The exchange sends the closing order to the order book. The order trades at the prevailing price, which can differ from the price at which the position became eligible for liquidation. Some exchanges also close a position in parts, and some use an insurance fund or other loss-sharing rules.

A liquidation record usually contains the exchange, the asset, the side, the size and the time. Providers collect these records and add them up by asset, by exchange and by time window.

These totals describe the past. A large value tells you that many forced closures were reported in that window. It does not tell you that the market has finished adjusting or that it will adjust further.

Long and short liquidations

A long position loses value when the price falls. A long liquidation therefore happens during a price decline. A short position loses value when the price rises. A short liquidation therefore happens during a price rise.

Read the two sides separately. A large long total and a small short total describe a window in which long positions were forced out. The same total split evenly describes a window with moves in both directions. The sum alone hides this difference.

A forced closure also pushes on the price. A long liquidation is a forced sell order. A short liquidation is a forced buy order. This is a mechanical fact about the order flow. It does not show which way the price will go next.

A long total and a short total are not symmetric in meaning. Markets rarely fall and rise at the same speed, and traders do not hold the same amount on both sides. A larger long total does not mean that more traders were wrong. It means that more long exposure reached its limit in that window.

The chart below shows the split between long and short liquidations in the public snapshot. Read each side before the total.

Exchange coverage and reporting gaps

No page covers every exchange. A total describes the covered exchanges only. A venue that is missing from the page can hold a large share of the real liquidations. Always find the list of covered exchanges before you quote a total.

Exchanges also report in different ways. Some publish every forced order, and some publish only part of them in their public feed. Some report the size in contracts, and some report it in coins or dollars. A provider has to convert these formats, and each conversion adds a possible error.

Because of these gaps, compare exchanges with care. A difference between two exchanges can be real, or it can come from the way each one reports. Use exchange comparisons to see where reported activity concentrated, not to rank exchanges.

Check the unit as well. A total in dollars reflects the price at the time of each closure. A total in coins removes the price. When you compare a calm day with a volatile day, the dollar value of the same coin amount can differ.

A liquidation total is a total of the covered exchanges, not of the whole market.

What a liquidation heatmap models

A heatmap shows price levels and a colour scale. Brighter areas mark levels where the model expects more leveraged positions to reach their liquidation price. The map is a picture of assumptions, not a list of real positions.

The model needs inputs that exchanges do not publish. It must guess where positions opened, how much leverage they used and what margin rules apply. Most models use open interest changes, price history and a set of common leverage levels to build this guess.

Two heatmaps can disagree for good reasons. They may use different leverage sets, different data sources or different rules for removing old positions. The colour scale is relative as well. A bright area means bright compared with the rest of that map, not bright in dollars.

A heatmap can also include positions that the owner already closed or hedged. It cannot see these changes. Read it as a picture of possible exposure under stated assumptions.

A bright area on a heatmap is a modelled concentration, not a confirmed cluster of positions.

Why two positions liquidate at different prices

Two positions opened at the same price can have very different liquidation prices. Leverage is the first reason. A position with more leverage has less margin for each unit of exposure, so a smaller price move removes the margin.

Margin mode is the second reason. In isolated margin, a position can lose only the margin assigned to it. In cross margin, the whole account balance backs the position, and other positions in the account change its liquidation price. A heatmap rarely knows which mode a trader chose.

Adding or removing collateral changes the level too. A trader who adds margin moves the liquidation price away from the market, and a trader who withdraws margin moves it closer. These changes are invisible in aggregated charts.

These facts explain why a modelled level is an approximation. They also explain why reported liquidations can occur at prices where a map shows no bright area.

Cascades as a description

A cascade is a sequence in which forced closures move the price and the price movement triggers more forced closures. The term describes something that happened. It is a label for a pattern in past data.

In liquidation data, a cascade appears as a cluster of reported liquidations close together in time, usually on one side. Open interest often falls during the same window because the closed contracts leave the market.

Do not use the word to predict. A heatmap area does not say that a cascade will start there. A past cascade does not say that the next move through the same level will repeat it. Many factors decide whether forced closures chain together, such as liquidity in the order book and how many positions were already closed.

Reading totals and heatmaps together

Use each source for the question it can answer. The reported data answers what was closed, on which side and on which exchange. The heatmap answers where a model places possible exposure. Combine them by asking questions after the fact.

• Did reported liquidations cluster near a level that the map marked?
• Did the clustered side match the side the map marked at that level?
• Did open interest fall in the same window?
• Did funding change after the window?

Comparing a model with reported events after the fact checks the model; it does not check the next event.

Time windows and outliers

Totals depend on the time window. A rolling window moves with the clock. A calendar window starts at a fixed boundary. The same event can sit in one window and outside another. Check how the page defines the window and which time zone it uses.

A single very large liquidation can dominate a total. In this case the total describes one event, not many. Look at the count of liquidations as well as the dollar amount when the page shows both.

Late reports can also change a window after the fact. A provider can add records that arrived late or remove duplicates. The newest window is therefore the least settled. Check it again later before you quote it.

Never compare a short window on one asset with a long window on another. Use the same asset and the same window for exchange comparisons, and keep the same window when you compare two days.

Questions