The Lab

How we tested

• Our own collected data, from Binance, January 2022 to 14 October 2026, for the 54 coins with at least six months of history before October 2025:
  • open interest: how many futures contracts are open;
  • the price;
  • funding: what one side pays the other every few hours to stay in;
  • long/short ratios: how many accounts, and how many top traders, lean long against short.
• Liquidation history doesn't exist in our data before 11 October 2026. So a flush was measured by a stand-in: within one hour, open interest falls more than its usual worst 1%, and the price moves more than its usual worst 1%. The five days of real liquidations were used only to check that stand-in.
• Ten kinds of event:
  • a flush down (longs forced out) or up (shorts forced out);
  • the same price move without the open-interest drop;
  • funding at either extreme;
  • either ratio at either extreme.
• What we measured: the price 1, 4, 24 and 72 hours after each event, compared with random times on the same coins. This is an "event study", a method from Fama, Fisher, Jensen and Roll (1969).
• Fair judging:
  • the extremes were set on January 2022 to September 2025;
  • results judged once on October 2025 to October 2026;
  • 40 tests, so the bar was raised to match (a Bonferroni correction);
  • to pass, an event also had to beat the 0.11% trading fee in the last year, on most coins.

The maths, step by step

1. A flush. Say a coin's open interest drops from 100,000 contracts to 97,000 in an hour (−3%) while its price falls from 80,000 to 78,500 (−1.9%). If that coin's worst 1% of hours are beyond −2.8% for open interest and −1.7% for the price, both are past their marks: a flush down.

2. Why bunched events fool a test. On 10 October 2025, the largest liquidation day on record, dozens of coins flushed in the same hours. If 30 coins bounce together the next day, a simple count sees 30 wins. But it was one market move, and it could easily have gone the other way. In the last year, the up-flushes were 108 events on only 53 days, and 30 of them were on 10–11 October 2025.

3. Counting each day once. Average each day's events into one number, then test the days. Three coins up 1%, 3% and 2% on the same day count as one day at +2%.

4. Taking the market out. If a coin rose 1.2% in the 24 hours after its flush and the whole market rose 1.0% over the same hours, only 1.2% − 1.0% = +0.2% was about that coin.

What we found

1. The first test passed 12 of 40, but the test was wrong.

• Our random comparison picked times independently for each coin, while real events bunch on crash days (step 2). That made the results look far stronger than they were.
• We caught this after the first look, and added two checks: each day counted once (step 3), and the market's move taken out (step 4). We say plainly that these came after the first look.

2. Counted by day, three results remained.

• After a sharp one-hour drop, the price rose more than usual over the next 24 to 72 hours. It rose 0.88% in 24 hours on average in the first years, and 1.17% in the last year (276 days, 65% of coins). An open-interest drop wasn't needed: the price drop alone did it.
• After the crowd leaned heavily long, the price fell over the next 72 hours, but that was in a last year when everything fell.

3. With the market taken out, nothing passed. The bounce after a sharp drop was the whole market bouncing, not the coin. (Short bounces after sharp falls were documented in stocks, by Jegadeesh and by Lehmann in 1990.) The crowded-long fall turned into a rise once the market's own fall was removed.

4. Forced selling added nothing. The flush (open interest falling with the price) did no better than the same drop without it. Funding extremes and the top traders' ratio showed nothing once counted by day.

5. The check on real liquidations (five days):

• the hours with the most liquidations (the top 1%) had open interest falling 93% of the time;
• but over all hours, liquidations and open-interest drops barely moved together;
• so the stand-in catches the biggest bursts and misses most others.

Update, 27 October 2026: where the idea comes from

Added after this page was published. The text above is left as it was.

• Forced sellers sell at bad prices. Andrei Shleifer and Robert Vishny described "fire sales" in 1992: when holders must sell at once, the natural buyers are often in the same trouble, so prices fall below value.
• Forced selling can feed itself. Markus Brunnermeier and Lasse Pedersen showed in 2009 how margin calls make traders sell, which moves the price, which brings more margin calls.
  • For example: a 2% fall liquidates 50x longs. Their forced sells push the price down another 1%, which reaches the 33x longs, and so on down.
• "The crowd is wrong." That a crowd leaning one way tends to lose is widely held among traders. In 1990 J. Bradford De Long, Andrei Shleifer, Lawrence Summers and Robert Waldmann showed how traders following the mood can push prices away from value for a while. Funding and long/short ratios are the usual way to see the crowd in crypto.
• The catch: crashes come in bunches. Many coins flush in the same hours, so a count of events overstates the evidence (step 2). That is what fooled this study's first test, and why each day is counted once.
• Theirs and ours. Fire sales, margin spirals and noise traders are theirs, and so are the event study, Bonferroni's guard and short-term reversal, all credited below. The flush stand-in, the ten events and the day-by-day count are ours.

Credits

• Eugene F. Fama, Lawrence Fisher, Michael C. Jensen and Richard Roll, "The Adjustment of Stock Prices to New Information", International Economic Review, 1969: the event study and the market-adjusted return.
• Carlo Emilio Bonferroni, Teoria statistica delle classi e calcolo delle probabilità, 1936, and Olive Jean Dunn, "Multiple Comparisons Among Means", Journal of the American Statistical Association, 1961: raising the bar for many tests.
• Narasimhan Jegadeesh, "Evidence of Predictable Behavior of Security Returns", Journal of Finance, 1990; Bruce N. Lehmann, "Fads, Martingales, and Market Efficiency", Quarterly Journal of Economics, 1990: short-term reversal.
• Andrei Shleifer and Robert W. Vishny, "Liquidation Values and Debt Capacity: A Market Equilibrium Approach", Journal of Finance, 1992: fire sales.
• Markus K. Brunnermeier and Lasse Heje Pedersen, "Market Liquidity and Funding Liquidity", Review of Financial Studies, 2009: forced selling that feeds itself.
• J. Bradford De Long, Andrei Shleifer, Lawrence H. Summers and Robert J. Waldmann, "Noise Trader Risk in Financial Markets", Journal of Political Economy, 1990: traders following the mood.
• Open interest, funding and long/short ratios are the exchanges' own published measures. The flush proxy and the tests are our own design.
• Data: Binance's public open interest, funding, ratios and prices, from our own collector.
• PerpGremlin and Claude, Forced selling and crowded trades: open interest, funding, long/short ratios, TerminalGremlin, 2026: the tests, the code and the words.