How to Read a Liquidation Cascade
There is no liquidation feed. There is a footprint, it is free, and it says three things almost nobody repeating the word actually believes.
By Pavel Penev, MScFounder, TradeWize · 10+ years trading the marketsThe short answer
A liquidation is an exit you did not choose: your margin ran out, and the exchange closed the position for you. A cascade is that happening to enough people at once that the forced selling moves the price and triggers the next round. Nobody publishes a history of them, so the way to find one is its footprint — a forced exit closes a position that already existed, so open interest FALLS. Price falling while open interest RISES is the opposite thing: new sellers arriving. Six years of bitcoin, ether and solana say that most violent hours are the second kind, not the first.
Every time crypto falls hard, the same word arrives within the hour. Cascade. Longs got flushed. A billion in liquidations. It is said with total confidence and almost never with a number, because the number is genuinely hard to get.
Here is the awkward part. There is no public history of liquidations. Binance's public archive publishes trades, klines, book depth and open interest — and no liquidation dataset. Bybit's has trading and index data and no liquidation folder either. What is left is a live feed with no past, and a handful of aggregators who will sell you their version of it. So "a billion got liquidated" is, almost always, a number nobody reading it can check.
There is a way around that, and it costs nothing. A forced exit CLOSES a position — it cannot open one. So it always removes open interest. That makes the footprint measurable in data anybody can download: watch price and the standing count of open contracts on the same clock, and the two of them together say which of two very different things just happened.
What a liquidation actually is
You put up margin to hold a leveraged position. The exchange watches that margin against the position's loss, and there is a level — the maintenance margin — below which it stops waiting. At that point it does not ask you. It sells your position into the market at whatever the market will pay.
That last detail is the whole mechanism. A forced exit is a market order that has to fill. It does not shop for a good price, it does not wait for a bounce, and it arrives at the exact moment everybody else's position is also underwater. So it pushes the price a little further in the direction that is already hurting — which pushes the next account past its own level, which sells again.
That is the cascade: not selling, but selling that manufactures the conditions for more selling. It is a feedback loop, and it stops for the boring reason that it runs out of fuel — there is nobody left close enough to their level to be next.
This is not a crypto invention
Margin calls have blown up markets for a century, and the mechanism is identical. What crypto changed is the speed and the leverage: the loop runs at machine speed, all day, every day, on positions sized at a level a stock broker would never permit. The result is that a cascade which would take a stock market a week takes a perpetual futures market about forty minutes — which is exactly what the clock below shows.
The footprint: two numbers on one clock
Open interest is the count of contracts currently open. It goes up when somebody opens a position and down when somebody closes one — and a forced exit is a close. That gives you a test with only two ingredients, both free.
Price down, open interest down
The flush. Leverage is leaving.
- Positions are being closed into weakness.
- The people who were already in are getting out — voluntarily or otherwise.
- A forced exit can only ever land in this column.
Price down, open interest up
Not a flush. Leverage is building.
- Positions are being opened into weakness.
- New sellers are arriving and taking the other side.
- Identical red candle, opposite state of the market.
Both look identical on a price chart. A red candle is a red candle. But they are opposite states of the market, and they happen about equally often — which is already at odds with how the word cascade gets used, and it gets worse when you look only at the violent hours.
| What happened | Share of hours | What it means |
|---|---|---|
| Price down, open interest down | 24.3% | Positions closing into weakness. The flush. |
| Price down, open interest up | 25.1% | New sellers arriving. Fresh bearish positioning, not a flush. |
| Price up, open interest down | 23.6% | Shorts closing into strength. The squeeze. |
| Price up, open interest up | 27.1% | New buyers arriving. Fresh leverage on the long side. |
629,701 rolling one-hour windows, Binance USDⓈ-M futures, 2020-09-01 to 2026-09-09.
Finding 1: most crashes are not cascades
Take the worst 0.1% of hours for price on each contract — the genuinely violent ones, a median of 4.5% on bitcoin in a single hour — and ask how many of them were actually forced closing.
Bitcoin is the most cascade-prone of the three and it still only manages 72.0%. Ether is 66.3%. Solana is 51.5% — a coin flip. Roughly one violent hour in three on bitcoin, and one in two on solana, is not a flush at all. It is people opening new positions into the fall, which is a market doing something ordinary rather than a market breaking.
That does not make the word wrong. Cascades are real, and the biggest ones are enormous. It makes the word over-applied: it is reached for every time price drops fast, and most of the time the data underneath it says something else entirely.
Finding 2: the clock you use decides the answer
Here is where it gets genuinely tricky, and where most readings of this go wrong. The same test, run over the same data, flips its answer depending on how long a window you measure.
In the worst 0.1% of windows for price, the share in which open interest also fell. 15 minutes against 4 hours.
On a 15-minute clock, solana's worst windows show open interest falling 61.7% of the time. Stretch the same measurement to 4 hours and it falls to 28.1% — and the median reading turns POSITIVE, at +8.54%. Four hours after solana's worst crashes, there is MORE leverage in the market than before them.
Nothing about the crashes changed. What changed is that a four-hour window is long enough to contain both the flush and the crowd that piles straight back in afterwards, and the second one is bigger. Measure over a long enough window and you erase the event you were trying to look at.
The practical version
If you want to know whether something was a flush, look at it on the shortest clock you have — minutes, not hours, and certainly not days. A daily candle almost never shows you a cascade. It shows you the cascade plus the recovery plus the re-entry, netted into one number that means none of the three.
Learn it by doing
Reading about it is one thing — it clicks when you do it. Learn it hands-on with free, interactive lessons on TradeWize.
Try the free lesson →Finding 3: nearly half of them happen while price is going up
The word cascade has a direction baked into it. Everybody pictures longs being carried out as price falls through the floor. So take the twenty largest hourly falls in open interest on each contract — the biggest forced closings in the record — and check which way price had been moving into them.
Read off the four hours into each event, not the hour itself — an hour that starts at a crash low shows a rise by construction.
28 of the 60 biggest flushes on record followed a four-hour RISE. Those are shorts being forced out, not longs — the squeeze, which is the same mechanism pointed the other way and gets a fraction of the attention. On ether the majority of the biggest events were short flushes, and on solana it is an exact split.
The single largest hourly fall in bitcoin open interest in six years is the cleanest example. On 26 July 2021, 22.6% of the entire standing position closed inside one hour — a bigger flush than 10 October 2025 — with price +5.77% during it and a +5.50% run-up into it. Nobody calls that day a cascade. It was the largest one on the record.
| Hour (UTC) | Open interest | The four hours into it | Price that hour | Side carried out |
|---|---|---|---|---|
| 2021-07-26 00:55 | −22.6% | +5.5% | +5.8% | shorts |
| 2025-10-10 21:20 | −17.7% | −7.9% | +3.7% | longs |
| 2020-11-26 02:45 | −16.0% | −1.1% | −4.4% | longs |
| 2020-09-03 22:45 | −14.8% | −0.2% | −5.4% | longs |
| 2023-04-26 19:15 | −14.3% | −0.6% | −6.3% | longs |
| 2024-04-13 19:40 | −13.6% | −1.5% | −6.4% | longs |
| 2024-01-03 11:50 | −13.3% | −1.6% | −4.2% | longs |
| 2024-02-28 16:35 | −13.1% | +4.0% | −2.0% | shorts |
| 2021-05-13 00:00 | −12.4% | −8.9% | +0.0% | longs |
| 2021-01-15 16:10 | −11.7% | −8.0% | +2.3% | longs |
Events within three days of a bigger one are dropped, so one crash does not fill the table.
The anatomy of 10 October 2025
The largest liquidation event crypto has had is worth putting under the fastest clock available, because the daily version of it — the one everybody has seen — hides the most interesting thing about it.
Bitcoin spent the whole day doing nothing. At 14:00 UTC it was $121,893 with 96,247 contracts open. Seven hours later, at 21:00, it was $115,007 — a slow bleed, and the standing position had barely moved.
Then, in about 10 minutes, price went to $101,516. That is 16.7% below where it had traded all afternoon, and the whole move happened between 21:10 and 21:20 UTC.
Five-minute readings. The shaded band is from the low at 21:20 to 22:00 — after the crash, during the recovery.
Now look at the bottom line, which is the part the daily row cannot show you. At the low, open interest was 89,433 — barely moved. The collapse in the standing position happens AFTERWARDS. Over the 40 minutes from the low to 22:00, 15.5% of every open contract closed, against 7.1% over the seven hours before it.
And price was going UP the whole time. By 22:00 bitcoin was back at $113,119, 11.4% off the low, and finished the day at $114,552. The hour containing the low lost 17.0% of the market's entire open position — most of it while the chart was recovering.
Why the flush lags the crash
The crash is what pushes accounts through their level. The forced closing is what happens next — the exchange working through the queue of positions that failed, filling them into whatever liquidity has returned. So the price low and the liquidation peak are not the same moment, and on this day they were about half an hour apart. Anybody watching only the chart saw a violent dip and a fast recovery. The actual event was still running.
Finding 4: none of this predicts anything
Which brings us to the part that would be easy to leave out. Everything above is a way of telling two states of the market apart. The obvious next question is whether the distinction is worth money — whether a crash with leverage leaving behaves differently afterwards from a crash with leverage building.
So: every hour where price fell 3% or worse, split by what open interest did, scored on the following 24 hours. Events inside a day of each other are collapsed so one bad week does not dominate.
| Contract | What open interest did | Events | Mean next 24h | Higher a day later |
|---|---|---|---|---|
| bitcoin | fell — forced closing | 118 | +2.40% | 73.7% |
| rose — new sellers | 49 | +2.69% | 61.2% | |
| ether | fell — forced closing | 159 | +1.48% | 65.4% |
| rose — new sellers | 55 | +3.54% | 70.9% | |
| solana | fell — forced closing | 255 | +2.88% | 71.0% |
| rose — new sellers | 109 | +3.35% | 68.8% |
Gross, in-sample, one venue. No fees, no funding, no slippage.
They are the same. Bitcoin bounces +2.40% on average after a flush and +2.69% after fresh selling. On ether and solana the "new sellers" group did BETTER than the flush group, which is the opposite of the folk version — buy the flush, the leverage is clean, the bottom is in.
What actually shows up is much duller and applies to both columns equally: after any hour that bad, the next day tends to be up a bit. That is a statement about violent hours in a market that rose over the sample, not a statement about liquidations.
So what is it good for?
Reading, not forecasting. Knowing that the last hour was forced closing tells you that leverage just left and the people who were positioned are now out — which changes how you read the next move, what a bounce is made of, and how much weight to put on the crowd that was there ten minutes ago. It does not tell you the direction, and this study is quite clear that anybody selling you the direction version has not checked.
How to read one yourself
- Put open interest on the chart, in contracts rather than dollars. In dollars it moves when price moves and tells you less than you think.
- Use a short clock. Five or fifteen minutes. On a daily candle the flush and the re-entry cancel out.
- When price drops hard, look at what open interest did over the same minutes. Falling with it is a flush. Rising is new positioning, and is not the same event.
- Check which way price came INTO it before you decide whose position got closed. Coming off a rally, you are looking at shorts.
- Expect the closing to finish after the price low, not at it. The low is what triggers the queue; the queue takes time to clear.
- Do not trade the label. The data says the distinction describes what happened and predicts nothing about what comes next.
What is a liquidation cascade?
A chain of forced exits. When a leveraged position's margin falls below the maintenance level, the exchange closes it with a market order that has to fill at whatever price is available. That pushes price further in the same direction, which pushes the next account below its own level. The loop continues until nobody is close enough to their level to be next. On 10 October 2025 the whole thing took about 40 minutes on bitcoin and closed 15.5% of every open contract.
How can I see liquidations if there is no public feed?
You measure the footprint instead. A forced exit closes a position that already existed, so it always removes open interest — it can never add any. Watch price and open interest on the same short clock: both falling together is forced closing, price falling while open interest rises is new sellers arriving. Open interest is free on every major venue's public data.
Does a liquidation cascade mark the bottom?
Not reliably, and this study finds no evidence for it. Comparing every hour where bitcoin fell 3% or worse, the next 24 hours averaged +2.40% when open interest fell and +2.69% when it rose — effectively identical. On ether and solana the crashes WITHOUT a flush did better. Violent hours tend to be followed by a bounce either way; the cascade part adds nothing.
Do cascades only happen when price falls?
No, and that is the most common mistake. Shorts get liquidated on the way up by exactly the same mechanism. Of the 60 biggest hourly falls in open interest across bitcoin, ether and solana, 28 followed a four-hour rise — those were shorts being carried out. The single largest one in bitcoin's record, on 26 July 2021, happened with price +5.77% during the hour.
Is falling open interest always a liquidation?
No. Open interest falls whenever anybody closes, and most closing is voluntary — taking profit, cutting a loss, going flat for the weekend. What a fall rules out is the opposite case, somebody opening a position. That is why the study reads it alongside price and on a short clock: a large, fast fall concentrated in minutes is a very different thing from a slow drift lower over a day.
How do I avoid being liquidated?
Use less leverage than the exchange allows, and size the position so the distance to your liquidation price is far wider than the market's ordinary noise. A liquidation is not a stop loss — it fills at whatever the market will pay in the worst possible conditions, and on a day like the one above that gap can be enormous. Our guide to leverage covers the arithmetic of the distance.
Method, and what this cannot tell you
There is no public record of individual liquidations, so this measures their footprint instead. A forced exit closes a position that already existed, so it removes open interest. Every rolling window is scored on two numbers at once: what price did, and what the standing count of open contracts did over the same minutes. Open interest is counted in CONTRACTS, not dollars, so it does not move just because price did. Windows whose five-minute readings are not contiguous are dropped rather than spanning a gap in the archive.
- One venue. Binance is the largest perpetual venue but not the market. A position force-closed on Bybit, OKX or Hyperliquid leaves no mark in here.
- Open interest is a footprint, not a liquidation feed. It falls when anybody closes — a forced exit and a voluntary one look identical in this data. What the number rules out is the opposite case, a position being opened.
- The readings are five-minute snapshots of an exchange-reported aggregate, so the moment a change is booked can trail the trade that caused it. One snapshot inside the 10 October window (21:45 UTC) prints far above both its neighbours and is an obvious bad reading; it is excluded from the anatomy and named here rather than quietly dropped.
- Two of the three contracts start in December 2021, because that is where Binance's public metrics archive starts for them. Only bitcoin has the full six years.
- In-sample and gross. No fees, no funding, no slippage, and no attempt to trade any of it.
Learn the mechanics, not just the word
Our crypto track builds the perpetual contract from scratch — margin, mark price, your own liquidation price, and why cross and isolated margin fail differently. The cascade is the last lesson of the stage, and by the time you reach it you can compute where your own level sits instead of finding out.