Open interest doesn't predict direction. We checked 136,000 hours.
Long buildup, short covering, long unwinding: we tested the open interest playbook on 136,000 hours of Binance BTC, ETH and SOL data. Direction: no. Volatility: yes.
Data window: 1 October 2020 to 15 September 2026 · Binance USDT-M perpetuals, BTC, ETH and SOL · last re-run 16 September 2026. Descriptive microstructure, not financial advice.
The playbook everyone repeats
Open a guide to open interest on almost any exchange, terminal or education site and you get the same table. It has four rows. Indian futures-and-options traders, who use it more than anyone, have names for the rows: long buildup, short buildup, short covering, long unwinding. Crypto borrowed the table and the confidence that came with it.
| Price | Open interest | What it is called | What the guides say happens next |
|---|---|---|---|
| Up | Up | Long buildup, "new longs" | "A strong bullish trend. New money is supporting the upward price movement." (Bookmap, 2024) · "Trend continuation higher" (Phemex Academy, 2026) |
| Up | Down | Short covering | "Signals a potential reversal. The rise in prices is not supported by new positions." (Bookmap) · "Rally may lack follow-through" (Phemex) |
| Down | Up | Short buildup, "new shorts" | "A strong bearish trend. New money is supporting the downward price movement." (Bookmap) · "Trend continuation lower" (Phemex) |
| Down | Down | Long unwinding, "long liquidation" | "Suggests a potential bottom or reversal." (Bookmap) · "Selling may exhaust soon" (Phemex) |
CME's own futures course puts the general version in one sentence: increasing open interest is typically a confirmation of the trend, decreasing open interest can be a signal that the trend is losing strength. TradeFundrr (August 2026) says the long buildup cell "builds the largest stack of leverage above current price" and that price-down, OI-down declines "are often the tail end of a liquidation event". Mudrex adds an honest caveat: "These aren't trading rules or guaranteed outcomes."
The descriptive half of the table is correct. If price rises while open interest rises, positions are being opened on net; if price rises while open interest falls, someone is closing. Our own market-state feed prints exactly that kind of label: at the time of writing the BTC drivers line reads "OI +0.2%/1h, price +0.3%, new longs". The question is the second half of the table, the part that turns a description of who did what into a forecast of what happens next.
That part is testable, and nobody had tested it on crypto perpetuals at scale. So we did.
How we tested it
Binance publishes five-minute snapshots of open interest for every USDT-margined perpetual, in coins and in dollars, going back to October 2020 for BTC and December 2021 for ETH and SOL. We aligned those snapshots to hourly closes: 52,125 hours with a valid open interest print for BTC, 41,945 for ETH, 41,825 for SOL. Prices are Binance 1-hour closes from the same archive.
For each hour we computed the price change and the open interest change over the previous 1, 4 and 24 hours, then the forward return over the next 4, 24, 72 and 168 hours. Open interest changes are measured in coins, not dollars. Measured in dollars, a change in open interest already contains the change in price, which makes the test circular. More on that below.
Three definitions of "a move", because the guides never say how big a move has to be. The sign version counts every hour. The main version requires both the price change and the open interest change to be above their own median absolute size; on the 24-hour window that means a price move of at least 1.37% and an open interest move of at least 2.09% for BTC. A strong version requires both to be in the top quarter. A fourth variant uses z-scores against a rolling 30-day standard deviation. All four agree, so the tables below use the main version.
One correction matters more than any other, and skipping it is how most studies of this kind go wrong. Consecutive hours inside the same 24-hour move are not independent observations. BTC spent 5,669 hours in the long-buildup cell on the 24-hour window; those hours collapse into 544 non-overlapping 24-hour episodes and 351 non-overlapping 72-hour episodes. Every statistic here is computed on the collapsed set, the same overlap rule we used in the funding extremes study.
We report medians rather than means, because crypto forward returns have fat tails, and "continuation": the share of episodes in which the forward return had the same sign as the move that put the hour into its cell.
Does the price and open interest combination predict direction?
No. Here is BTC on the 24-hour window, both moves above median size, forward 24 and 72 hours.
| Cell | The guides say | Episodes (24h) | Median next 24h | Continuation, 24h | Episodes (72h) | Median next 72h | Continuation, 72h |
|---|---|---|---|---|---|---|---|
| Price up, OI up (long buildup) | trend continues | 544 | +0.09% | 51.1% | 351 | +0.40% | 51.9% |
| Price up, OI down (short covering) | weak rally, reversal | 419 | −0.12% | 47.5% | 305 | +0.36% | 52.5% |
| Price down, OI up (short buildup) | trend continues | 395 | +0.18% | 45.8% | 277 | +0.44% | 45.1% |
| Price down, OI down (long unwinding) | bottom near | 474 | +0.33% | 44.3% | 343 | +0.49% | 44.9% |
| All hours (baseline) | 2,176 | +0.02% | 725 | +0.25% |
Continuation rate over the next 72 hours after each price and open interest combination. 24-hour window, both moves above median size, non-overlapping episodes. Dashed line: 50%. Hover or tab across the bars for the 24-hour numbers.
Hover a bar for the 24-hour and 72-hour numbers.
Table view
| Cell | n (24h) | Median 24h | Cont. 24h | n (72h) | Median 72h | Cont. 72h |
|---|---|---|---|---|---|---|
| BTC: Price up, OI up (long buildup) | 544 | +0.09% | 51.1% | 351 | +0.40% | 51.9% |
| BTC: Price up, OI down (short covering) | 419 | −0.12% | 47.5% | 305 | +0.36% | 52.5% |
| BTC: Price down, OI up (short buildup) | 395 | +0.18% | 45.8% | 277 | +0.44% | 45.1% |
| BTC: Price down, OI down (long unwinding) | 474 | +0.33% | 44.3% | 343 | +0.49% | 44.9% |
| ETH: Price up, OI up (long buildup) | 436 | −0.20% | 47.5% | 279 | −0.33% | 47.3% |
| ETH: Price up, OI down (short covering) | 276 | −0.35% | 46.0% | 205 | +0.27% | 52.7% |
| ETH: Price down, OI up (short buildup) | 304 | +0.21% | 46.7% | 207 | +0.37% | 45.9% |
| ETH: Price down, OI down (long unwinding) | 412 | +0.04% | 49.0% | 275 | −0.02% | 50.5% |
| SOL: Price up, OI up (long buildup) | 373 | −0.41% | 48.5% | 228 | −0.38% | 49.1% |
| SOL: Price up, OI down (short covering) | 325 | −0.34% | 46.5% | 237 | −1.29% | 43.0% |
| SOL: Price down, OI up (short buildup) | 327 | +0.09% | 48.9% | 224 | +0.56% | 47.8% |
| SOL: Price down, OI down (long unwinding) | 391 | 0.00% | 49.9% | 267 | +0.74% | 47.2% |
Read the two "price up" rows first. After a long buildup, BTC kept rising over the next 72 hours in 51.9% of episodes. After a short-covering rally, the supposedly weak one, it kept rising in 52.5%. The open interest column moved the answer by 0.6 percentage points, in the wrong direction.
Now the two "price down" rows. After new shorts came in, the decline continued 45.1% of the time. After long unwinding, 44.9%. Both cells resolved up about 55% of the time, which is not a signal about open interest. It is BTC drifting upward over six years, plus a little mean reversion after any down day.
ETH and SOL tell the same story with different noise. On ETH, long buildup continued 47.3% of the time at 72 hours and short covering 52.7%; on SOL, 49.1% and 43.0%. On the down side, ETH's short buildup continued 45.9% and long unwinding 50.5%; SOL's, 47.8% and 47.2%. Twelve OI-up-versus-OI-down contrasts across three assets, two windows and two horizons, and the smallest p-value in the set is 0.059 (SOL, 4-hour window, 4-hour horizon), which reverses to 0.42 at the next horizon. Bootstrap intervals on the difference in median forward return all contain zero; the widest for BTC at 72 hours runs from −0.80 to +0.91 percentage points around a point estimate of +0.03.
The sign of the effect also flips from year to year. On BTC, the gap in 72-hour continuation between long buildup and short covering was −2.5 points in late 2020, −3.9 in 2021, +17.6 in 2022, −7.7 in 2023, −1.9 in 2024, −8.1 in 2025 and −0.2 so far in 2026. One year in seven agreed with the guides. Five disagreed.
The daily-bar version, since that is how people look at it
Most traders read open interest on a daily chart: yesterday's price bar next to yesterday's open interest bar. So we ran the table on 00:00 UTC daily snapshots too.
| Cell | BTC days | BTC next-day continuation | ETH | SOL |
|---|---|---|---|---|
| Price up, OI up (long buildup) | 601 | 47.3% | 46.9% | 49.0% |
| Price up, OI down (short covering) | 496 | 50.4% | 47.7% | 47.8% |
| Price down, OI up (short buildup) | 511 | 48.7% | 45.9% | 51.1% |
| Price down, OI down (long unwinding) | 563 | 46.5% | 47.9% | 50.5% |
| Momentum only: up day followed by up day | 1,097 | 48.7% | 47.3% | 48.4% |
On BTC the long-buildup day, the one described as the healthiest thing a trend can do, had the lowest next-day continuation of the four cells, below even the plain "it went up yesterday" baseline. At three days out the four cells sit between 44% and 51%. The table does not help on daily bars either.
Dollar open interest is mostly price
Before the volatility result, a detour that explains a lot of open interest commentary.
Exchanges and dashboards quote open interest in dollars. Dollar open interest is coins times price, so when price moves 5%, dollar open interest moves 5% with no position opened or closed. Over 24-hour windows the correlation between the change in Binance BTC dollar open interest and the change in price is 0.67; for the change in coin open interest it is 0.19. On ETH the pair is 0.74 versus 0.24, on SOL 0.66 versus −0.02. In 17% of BTC 24-hour windows (20% on ETH, 24% on SOL) the dollar figure and the coin figure move in opposite directions.
The record headlines are where this bites. Since 1 January 2023, Binance BTCUSDT open interest has set a new all-time high in dollars on 34 separate days, 27 of them in 2024 and 7 in 2025, the last on 7 October 2025 at $12.49 billion. The number of new all-time highs in coins over the same period is zero. In coins, open interest on the largest perpetual venue peaked on 27 October 2022 at 166,691 BTC, worth $3.46 billion at the time, and it has never been back. On 14 September 2026 it stood at 103,517 BTC, 62% of the peak, while the dollar figure of $8.09 billion looked like 65% of a much more recent record. SOL is the same shape: a coin peak on 10 November 2022 and 54% of it today, against dollar records set in 2025.
Daily 00:00 UTC snapshots. Every record open interest day since 2023 was a dollar record; in coins, open interest peaked in October 2022. Hover or use ←/→ to scrub, click to pin.
- Date
- 2026-09-15
- Open interest
- 103,517 BTC
- In dollars
- $8.09B
- Price
- $78,153
Table view (monthly)
| Month (first snapshot) | OI, BTC | OI, $B | Price |
|---|---|---|---|
| 2020-10 | 35,654 | 0.38 | $10,773 |
| 2020-11 | 48,976 | 0.67 | $13,784 |
| 2020-12 | 38,904 | 0.77 | $19,720 |
| 2021-01 | 34,936 | 1.01 | $28,952 |
| 2021-02 | 31,301 | 1.04 | $33,126 |
| 2021-03 | 29,548 | 1.34 | $45,163 |
| 2021-04 | 34,381 | 2.02 | $58,807 |
| 2021-05 | 31,927 | 1.84 | $57,684 |
| 2021-06 | 37,822 | 1.41 | $37,243 |
| 2021-07 | 55,181 | 1.93 | $35,031 |
| 2021-08 | 54,187 | 2.25 | $41,448 |
| 2021-09 | 60,692 | 2.86 | $47,150 |
| 2021-10 | 47,921 | 2.10 | $43,797 |
| 2021-11 | 55,293 | 3.39 | $61,349 |
| 2021-12 | 59,230 | 3.37 | $56,975 |
| 2022-01 | 74,803 | 3.46 | $46,211 |
| 2022-02 | 82,922 | 3.19 | $38,439 |
| 2022-03 | 69,678 | 3.01 | $43,155 |
| 2022-04 | 75,200 | 3.42 | $45,506 |
| 2022-05 | 85,953 | 3.23 | $37,614 |
| 2022-06 | 87,997 | 2.80 | $31,798 |
| 2022-07 | 93,485 | 1.86 | $19,924 |
| 2022-08 | 113,426 | 2.64 | $23,290 |
| 2022-09 | 112,589 | 2.26 | $20,042 |
| 2022-10 | 125,456 | 2.44 | $19,412 |
| 2022-11 | 134,356 | 2.75 | $20,482 |
| 2022-12 | 127,166 | 2.18 | $17,153 |
| 2023-01 | 105,693 | 1.75 | $16,538 |
| 2023-02 | 97,523 | 2.25 | $23,119 |
| 2023-03 | 103,272 | 2.39 | $23,130 |
| 2023-04 | 107,670 | 3.06 | $28,455 |
| 2023-05 | 98,019 | 2.86 | $29,223 |
| 2023-06 | 111,646 | 3.04 | $27,201 |
| 2023-07 | 105,972 | 3.23 | $30,460 |
| 2023-08 | 108,877 | 3.18 | $29,221 |
| 2023-09 | 82,466 | 2.14 | $25,928 |
| 2023-10 | 90,183 | 2.43 | $26,951 |
| 2023-11 | 88,244 | 3.06 | $34,651 |
| 2023-12 | 82,614 | 3.12 | $37,717 |
| 2024-01 | 74,006 | 3.13 | $42,314 |
| 2024-02 | 71,956 | 3.06 | $42,560 |
| 2024-03 | 71,955 | 4.40 | $61,203 |
| 2024-04 | 82,767 | 5.90 | $71,363 |
| 2024-05 | 72,607 | 4.40 | $60,651 |
| 2024-06 | 76,538 | 5.18 | $67,578 |
| 2024-07 | 85,707 | 5.38 | $62,766 |
| 2024-08 | 82,638 | 5.34 | $64,602 |
| 2024-09 | 80,995 | 4.77 | $58,942 |
| 2024-10 | 87,004 | 5.51 | $63,309 |
| 2024-11 | 89,576 | 6.30 | $70,322 |
| 2024-12 | 86,882 | 8.39 | $96,475 |
| 2025-01 | 91,253 | 8.55 | $93,549 |
| 2025-02 | 83,373 | 8.53 | $102,380 |
| 2025-03 | 78,949 | 6.65 | $84,300 |
| 2025-04 | 68,882 | 5.69 | $82,518 |
| 2025-05 | 80,107 | 7.55 | $94,125 |
| 2025-06 | 83,624 | 8.74 | $104,545 |
| 2025-07 | 76,382 | 8.18 | $107,087 |
| 2025-08 | 90,258 | 10.44 | $115,697 |
| 2025-09 | 89,854 | 9.71 | $108,208 |
| 2025-10 | 88,420 | 10.09 | $113,989 |
| 2025-11 | 78,320 | 8.57 | $109,557 |
| 2025-12 | 89,832 | 8.06 | $90,321 |
| 2026-01 | 94,262 | 8.26 | $87,608 |
| 2026-02 | 93,560 | 7.38 | $78,707 |
| 2026-03 | 77,330 | 5.18 | $66,937 |
| 2026-04 | 89,202 | 6.08 | $68,242 |
| 2026-05 | 95,352 | 7.29 | $76,305 |
| 2026-06 | 105,009 | 7.75 | $73,653 |
| 2026-07 | 108,532 | 6.37 | $58,605 |
| 2026-08 | 109,490 | 6.89 | $62,860 |
| 2026-09 | 107,897 | 8.48 | $78,550 |
ETH is the exception, and it shows why the distinction matters. ETH open interest made a fresh high in coins on 17 March 2026 at 2,475,057 ETH and sits at 94% of it. That is real positioning growth. The BTC and SOL "records" were price.
The rest of this study uses coins. Any open interest chart you read should too. For a live example, our event study of the week of 14 September 2026 counts the Clarity vote, the Fed hike and the break above $80,000 in coins, and finds a build where the dollar figure and the headlines saw a squeeze.
What open interest does predict: the size of the next move
Group every BTC hour by the size and sign of its 24-hour change in open interest, ten equal deciles from the biggest flush to the biggest build, and look at the high-to-low range of the following 24 hours.
BTC. Median high-to-low range of the next 24 hours, by decile of the 24-hour change in open interest (coins). Left: the biggest flushes. Right: the biggest builds. Hover or tab across the bars for the decile bounds.
Hover a bar for the decile bounds and sample size.
Table view (BTC, ETH, SOL)
| Decile | 24h OI change | Median next-24h range | n hours |
|---|---|---|---|
| BTC D1 | −45.7% to −4.3% | 4.32% | 5,204 |
| BTC D2 | −4.3% to −2.5% | 3.79% | 5,203 |
| BTC D3 | −2.5% to −1.4% | 3.44% | 5,203 |
| BTC D4 | −1.4% to −0.6% | 3.35% | 5,203 |
| BTC D5 | −0.6% to 0.1% | 3.43% | 5,203 |
| BTC D6 | 0.1% to 0.9% | 3.51% | 5,203 |
| BTC D7 | 0.9% to 1.8% | 3.67% | 5,203 |
| BTC D8 | 1.8% to 2.9% | 3.77% | 5,203 |
| BTC D9 | 2.9% to 4.9% | 4.07% | 5,203 |
| BTC D10 | 4.9% to 25.4% | 4.85% | 5,203 |
| ETH D1 | −28.8% to −4.2% | 4.79% | 4,189 |
| ETH D2 | −4.2% to −2.5% | 4.81% | 4,189 |
| ETH D3 | −2.5% to −1.4% | 4.46% | 4,189 |
| ETH D4 | −1.4% to −0.6% | 4.26% | 4,189 |
| ETH D5 | −0.6% to 0.0% | 4.35% | 4,189 |
| ETH D6 | 0.0% to 0.7% | 4.39% | 4,188 |
| ETH D7 | 0.7% to 1.6% | 4.47% | 4,189 |
| ETH D8 | 1.6% to 2.7% | 4.72% | 4,189 |
| ETH D9 | 2.7% to 4.6% | 4.70% | 4,189 |
| ETH D10 | 4.6% to 42.5% | 4.93% | 4,189 |
| SOL D1 | −45.1% to −5.4% | 7.42% | 4,177 |
| SOL D2 | −5.4% to −3.2% | 6.37% | 4,177 |
| SOL D3 | −3.2% to −1.9% | 6.37% | 4,177 |
| SOL D4 | −1.9% to −0.9% | 5.99% | 4,177 |
| SOL D5 | −0.9% to −0.0% | 5.86% | 4,177 |
| SOL D6 | −0.0% to 0.8% | 5.81% | 4,177 |
| SOL D7 | 0.8% to 1.9% | 6.19% | 4,177 |
| SOL D8 | 1.9% to 3.3% | 6.39% | 4,177 |
| SOL D9 | 3.3% to 6.0% | 7.03% | 4,177 |
| SOL D10 | 6.0% to 152.3% | 7.94% | 4,177 |
| 24h change in OI (coins) | Median range, next 24h |
|---|---|
| Decile 1, flush of 4.3% or more | 4.32% |
| Decile 2 | 3.79% |
| Decile 3 | 3.44% |
| Decile 4 | 3.35% |
| Decile 5 | 3.43% |
| Decile 6 | 3.51% |
| Decile 7 | 3.67% |
| Decile 8 | 3.77% |
| Decile 9 | 4.07% |
| Decile 10, build of 4.9% or more | 4.85% |
The shape is a U. After the biggest builds the next day's range was 45% wider than after the calmest decile; after the biggest flushes, 29% wider. ETH and SOL show the same U, shallower on ETH (4.26% in the middle, 4.79% and 4.93% at the ends) and steeper on SOL (5.81% in the middle, 7.42% and 7.94% at the ends).
Part of that is just volatility clustering: big open interest moves happen on volatile days, and volatile days are followed by volatile days. So we controlled for it, regressing the log of the next day's range on the log of the previous day's range, the absolute price change, the absolute open interest change and the signed open interest change, with errors corrected for overlap.
Two things fall out. The absolute size of the open interest move adds nothing once you know yesterday's range; on ETH the coefficient is negative. The sign does. A build in open interest predicts a wider next day and a flush predicts a narrower one, on all three assets, with test statistics of 6.5 on BTC, 2.6 on ETH and 2.7 on SOL. A 10-percentage-point build over 24 hours multiplies the next day's expected range by 1.16 on BTC, 1.07 on ETH and 1.05 on SOL. Without the regression: inside every quintile of past volatility, BTC's next-day range after a top-decile build was 16% to 34% wider than after a bottom-decile flush.
The mechanism is not mysterious. Open interest is leverage that has not yet been resolved. When it builds, there is more of it to resolve; when it flushes, the fuel has already burned. Open interest tells you how much is loaded, not which way it fires.
Two honest caveats. The effect is strong on BTC and modest on ETH and SOL. And it is a statistical result, not a trade: a 16% wider expected range does not pay for the spread and funding of a volatility position.
Does an open interest flush mark the bottom?
This is the one piece of the playbook with something behind it, so it deserves care.
A 24-hour drop in open interest of 5% or more predicted nothing: forward hit rates of 52% to 53% across the three assets, indistinguishable from baseline. At 8% or more, a lean appears. Pooling BTC, ETH and SOL, the 367 non-overlapping 72-hour windows that followed such a flush closed up 57.2% of the time with a median return of +0.92%, against roughly 52% for an average window; two-sided p = 0.045. At 10% or more, the seven-day window after the flush closed up 59.0% of the time with a median of +1.51% (n = 173, p = 0.067). On BTC alone the numbers are stronger and the sample smaller: after a 10% flush, 65.8% of 72-hour windows closed up (n = 73, median +1.70%) and 68.3% of 7-day windows (n = 60, median +2.11%).
Why we call it a lean. We tested three thresholds, three horizons and three assets, twenty-seven combinations, and a p-value near 0.05 is what you expect one or two of those to produce by chance. The direction agrees across assets at 8% and 72 hours, which is the only reason it is in the article. On SOL at 8% the seven-day median is negative. On ETH at 10% the 24-hour and 72-hour windows show nothing.
The twelve largest daily flushes in Binance BTC open interest since October 2020, with what followed:
| When (UTC) | OI change, 24h | Price change, 24h | Price | Next 24h | Next 72h | Next 7 days |
|---|---|---|---|---|---|---|
| 27 Nov 2020, 03:00 | −24.1% | −7.2% | $17,102 | −0.6% | +8.2% | +12.7% |
| 24 Dec 2020, 03:00 | −19.1% | −3.5% | $22,806 | +3.0% | +16.6% | +26.1% |
| 4 Jan 2021, 20:00 | −19.5% | −3.5% | $31,689 | +6.5% | +23.5% | +1.9% |
| 22 Jan 2021, 10:00 | −19.1% | −4.5% | $31,438 | +3.3% | +5.1% | +17.8% |
| 23 Feb 2021, 14:00 | −23.3% | −6.6% | $48,769 | +1.5% | −5.7% | +0.5% |
| 19 Apr 2021, 02:00 | −22.4% | −4.3% | $56,744 | −3.4% | −5.0% | −8.5% |
| 20 May 2021, 08:00 | −45.7% | −1.1% | $39,947 | −0.3% | −10.1% | −4.2% |
| 26 Jul 2021, 23:00 | −26.7% | +8.5% | $37,570 | +4.2% | +5.9% | +4.9% |
| 7 Sep 2021, 18:00 | −30.3% | −9.8% | $46,628 | −0.9% | −1.9% | +0.2% |
| 4 Dec 2021, 20:00 | −38.2% | −8.6% | $49,247 | −0.7% | +3.3% | −1.7% |
| 14 Jan 2023, 20:00 | −20.9% | +7.5% | $20,797 | +0.4% | +2.6% | +11.9% |
| 11 Oct 2025, 05:00 | −25.9% | −6.8% | $112,905 | −1.3% | +0.1% | −5.8% |
Eight of twelve up after 72 hours, eight of twelve up after seven days. The largest flush in the sample, 45.7% of open interest gone in the 24 hours around 19 and 20 May 2021, was followed by a further 10.1% decline over three days. Two of the twelve are short squeezes rather than long liquidations, 26 July 2021 and 14 January 2023, where open interest collapsed because shorts were forced out and price rose 8%; the August 2026 squeeze had the same shape. The table also shows how the market has changed size: after 2021, a 19% daily flush happened twice in five years. The same dollar cascade now moves a much smaller share of a much larger book, which is one more reason to measure flushes as a percentage of coin open interest rather than in billions of dollars of liquidations.
Open interest vs volume
Volume is flow: every contract traded during the day, counted every time it changes hands. Open interest is stock: the contracts still open at the end of the day. The glossary entry covers the definitions; what the archive adds is how the two compare in practice, and the ratio has changed more than most people think.
| Median daily volume divided by open interest | 2020 | 2021 | 2022 | 2023 | 2024 | 2025 | 2026 |
|---|---|---|---|---|---|---|---|
| BTC | 7.3× | 8.0× | 4.4× | 3.6× | 3.0× | 1.8× | 1.5× |
| ETH | 5.1× | 4.7× | 3.4× | 2.7× | 2.7× | 1.9× | |
| SOL | 5.1× | 6.5× | 5.7× | 3.4× | 2.7× | 2.2× |
In 2021 the Binance BTC perpetual traded eight times its open interest every day. In 2026 it trades one and a half times. Some of that is Binance losing share to other venues and a different fee and market-making structure, so we would not read it as "positions live longer" without more evidence. But it is the number to have in mind when someone puts a volume spike and an open interest spike in the same sentence: on today's book, a day of volume equal to open interest is an ordinary day.
Adding volume to the four-cell table does not rescue it, either. The classic three-way version (price up, volume up, open interest up equals strength) produced 16 cells per asset, and all of them sat between 40% and 55% continuation. The bullish flagship, price up with above-median volume and rising open interest, continued 50.8% of the time on BTC over 72 hours (n = 262).
What else we checked
Builds of 8% to 10% of open interest in a day, the "leverage piling in" scenario, produced positive 72-hour medians on BTC and ETH and nothing on SOL, and reversed sign at seven days. Open interest at a 90-day high, the "crowded" scenario, was followed by a −1.30% median 72-hour return and a 40.7% hit rate on BTC (n = 59), the opposite on ETH (+2.13%, 60%) and nothing usable on SOL (n = 40). Rallies in the top quintile of seven-day returns did slightly worse on BTC when open interest had grown with them (−0.45% median over 72 hours, 46.6% up) than when it had not (+0.38%, 52.8%), but ETH ran the other way. Trend framing, price above its 30-day average with open interest rising or falling over a week, gave the one result with the textbook's sign reversed: on BTC, uptrends that were shedding open interest returned +0.92% over 72 hours with a 58.5% hit rate, against +0.04% and 50.8% for uptrends adding it. ETH and SOL did not repeat it. None of these survive the consistency test we set at the start, so none of them are findings.
What this doesn't tell you
It is one venue. Binance is the largest perpetual market, 47% of cross-venue BTC open interest at the time of writing, but a cross-venue history in coins only exists in our archive from July 2026. The record-high finding is about Binance BTCUSDT and is stated that way.
It is one regime. October 2020 to September 2026 is one halving cycle, the ETF era and a couple of macro shocks. The open interest playbook was written for commodity futures decades before crypto existed; whether it worked there is a different study.
Thresholds are choices. We showed four ways of defining a move and a daily-bar version, and they agree, but a fifth definition always exists.
Medians hide tails on purpose. The 10th-to-90th percentile band of 72-hour forward returns in these cells runs about ±5% to ±7% on BTC.
We ran roughly ninety tests. Under that many, a p-value of 0.05 is weak evidence, which is why the flush result is a lean and the quadrant result, twelve contrasts with nothing below 0.059, is a null.
And the five-minute snapshots do not see what happens inside five minutes. That does not affect 24-hour tests. It does mean the "largest one-hour flush" is a lower bound.
Where the tape is now
Snapshot 16 September 2026, 10:03 UTC, before the Fed decision.
BTC at $76,000. Cross-venue open interest $17.44 billion: Binance $8.13 billion, Bybit $4.37 billion, Hyperliquid $2.78 billion, OKX $2.15 billion. Since 9 September, price has gone from $79.7k to $76.0k, about −5%, while cross-venue open interest went from $17.9 billion to $17.4 billion, about −3%. The playbook calls that long unwinding and says a bottom is near. The data above says it is a cell with 45% continuation and a coin flip either way. Funding sits at the 46th percentile of its two-year history, and the drivers line on the positioning page currently reads "OI +0.2%/1h, price +0.3%, new longs", which is an accurate description of the last hour and, as far as we can measure, no description of the next one.
Check it yourself
The live open interest, its change over the last hour and its level against the monthly median are on the positioning view for all seven assets, in dollars and per venue. The conditional outcomes engine in the agent feed accepts open interest change as a condition (oi_chg_1h_pct), so any base rate in this piece can be re-derived, and any condition we did not test can be tried, against a 15-minute cross-venue archive that grows every day. The archive is young, so the sample sizes it returns are small for now; it says so in the response.
Related
→ Extreme funding rates: what 789 days of data actually show
→ Anatomy of a short squeeze: Bitcoin's 45-hour run to $79.6K
→ Liquidity grabs and stop hunts: the real part and the myth
→ Long/short ratio versus OBI and CVD
→ How to read a liquidation heatmap
Watch the same data live and free: Positioning for open interest, funding and the drivers line, Liquidations for the forced flow that resolves it.
How to cite this research
MarketTrace (2026). "Open interest doesn't predict direction. We checked 136,000 hours." Price and open interest quadrant study on Binance USDT-M perpetuals, BTC 2020-10-01 to 2026-09-15, ETH and SOL 2021-12-01 to 2026-09-15. https://markettrace.ai/blog/open-interest-analysis. Figures are re-run quarterly; the dateModified stamp reflects the latest run.
Frequently asked questions
What does open interest indicate?
Open interest indicates how many perpetual contracts are open, which is the amount of leverage that has not yet been closed or liquidated. Measured on 136,000 hours of Binance BTC, ETH and SOL data (2020 to 2026), a change in open interest did not indicate the direction of the next move: after each of the four price-and-open-interest combinations, price continued in the same direction 43% to 53% of the time. It did indicate the size of the next move: a 10-percentage-point build in open interest over 24 hours widened the next day's range by 16% on BTC after controlling for recent volatility.
What are long buildup, short buildup, short covering and long unwinding?
They are the four combinations of price direction and open interest direction. Long buildup is price up with open interest up (new longs opening). Short buildup is price down with open interest up (new shorts opening). Short covering is price up with open interest down (shorts closing). Long unwinding is price down with open interest down (longs closing or being liquidated). The labels describe what happened to positions and are accurate as descriptions. As forecasts of the next move, none of the four had a measurable edge on Binance BTC, ETH or SOL perpetuals between 2020 and 2026.
Does rising open interest confirm a trend?
Not in this sample. After a 24-hour price rise with rising open interest, BTC continued higher over the next 72 hours in 51.9% of 351 non-overlapping episodes; after a price rise with falling open interest, 52.5% of 305. On daily bars, price-up-with-OI-up days were followed by another up day 47.3% of the time, the lowest of the four combinations. ETH and SOL showed the same absence of a gap, and the sign of the small differences flipped from year to year.
Is open interest measured in coins or dollars?
Both, and it matters which. Dollar open interest equals coin open interest times price, so it rises and falls with price even when no position changes; its 24-hour change correlates 0.67 with the BTC price change, against 0.19 for the coin figure. Every "record open interest" day on Binance BTCUSDT since January 2023 (34 of them) was a dollar record; in coins, open interest peaked on 27 October 2022 at 166,691 BTC and is at 62% of that level in September 2026. Analysis of position flow should use coins.
What counts as a big change in open interest?
On Binance BTCUSDT, 2020 to 2026, the median absolute 24-hour change in coin open interest is 2.1%. A change of 6.6% is a top-decile day and 14% is a top-percentile day. Over one hour, the median absolute change is 0.3%, a 1.2% move is top-decile and 3.3% is top-percentile. Flushes of 10% or more in 24 hours happened 73 times in six years on BTC, counted as separate episodes at least 72 hours apart; flushes of 20% or more, nine times.
Does a drop in open interest mean the bottom is in?
Weakly, and only for large drops. Across BTC, ETH and SOL, a 24-hour drop in open interest of 8% or more was followed by a positive 72-hour return 57% of the time with a median of +0.92% (367 episodes, p = 0.045); drops of 5% carried no information. Of the twelve largest BTC flushes since 2020, eight resolved higher after three days and four resolved lower, including the largest one, on 20 May 2021, which was followed by a further 10% decline.
MarketTrace shows aggregated open interest, funding, order flow and liquidations for BTC, ETH, SOL, BNB, XRP, DOGE and HYPE across Binance, Bybit, OKX and Hyperliquid. Informational data feed only. Not financial guidance.