The ETH Share of Crypto Leverage Is a Venue Statistic, Not a Market One: 27.99 and 42.44 Percent on the Same Day

"Leverage is rotating into Ether" is a claim people make with a number attached. The number is usually the ETH share of open interest: Ether's open interest divided by Bitcoin's plus Ether's, read off whichever dashboard was open. It is a reasonable thing to want. If the share is climbing, more of the market's committed leverage is sitting in ETH, which changes where a liquidation cascade starts.

I measured that share on three large venues, daily: 2,184 days on Bybit back to 2020-10-22, 180 on OKX and the 30 that Binance serves on its history endpoint. The answer depends on a horizon and on a venue, and the second dependence is the one nobody prices in. Over a 200-day horizon the disagreement between venues about what the share is today is about as large as the entire rotation across those 200 days. Over a multi-year horizon the rotation wins comfortably. What holds across every day I can compare the two on is that two venues measuring the same two contracts disagree by roughly 12 points and never cross, though that comparison is limited to the 30 days Binance serves.

What share of crypto perpetual leverage sits in Ether rather than Bitcoin?

On 2026-10-14 the answer is 27.99 percent, 35.35 percent or 42.44 percent, depending on whose book you read. Bybit's USDT-margined perpetuals put Ether at 27.99 percent of the combined BTC plus ETH dollar open interest; OKX's BTC and ETH contract series put it at 35.35 percent; Binance's USDT-margined perpetuals put it at 42.44 percent. Summing all three into one pool, on a consistent one-side convention, gives 38.63 percent, from 13.31 billion USD of Bitcoin open interest against 8.38 billion of Ether. Those figures were pulled at 16:39 UTC on 2026-10-14. The 14.44 percentage point spread between the lowest and the highest venue reading is not a data error and does not average away over time: across the 30 days to 2026-10-14 the Binance reading exceeded the Bybit reading on all 30 days, by a median of 11.93 points, and the two venues' 30-day ranges do not overlap at any point. Set against the rotation it is a question of horizon, and the honest version needs all three numbers. Over the 200 most recent Bybit days the entire rotation, trough to peak, is 15.28 points, so the venue gap and 200 days of rotation are the same size. Over 180 days of OKX history the rotation is 7.88 points, less than picking a different exchange moves the number on a single afternoon. But over Bybit's full published history, 2,184 matched days back to 2020-10-22, the share has run from 1.13 to 44.44 percent, and from 13.22 to 44.44 percent if the thin 2020 launch period is dropped. On a multi-year horizon the rotation is twice the venue gap and the comparison reverses.

Method, in six steps

1. Pulled daily open-interest history from three public, no-key endpoints: Bybit v5/market/open-interest (BTCUSDT and ETHUSDT, category linear, intervalTime 1d), Binance futures/data/openInterestHist (BTCUSDT and ETHUSDT, period 1d), and OKX v5/rubik/stat/contracts/open-interest-volume (ccy BTC and ETH, period 1D). The three endpoints are not equally deep and the difference is a trap worth naming, because two of the three limits are hard and one is not. Binance returns 30 rows where 500 were requested: period 1d with limit 500 still returns 30. OKX floors at 2026-04-18 and refuses older ranges. Bybit's limit maxes at 200, and 200 is the page size, not the history: the response carries a nextPageCursor, and paginating it returns 2,262 BTCUSDT days back to 2020-08-05, of which 2,184 have an ETHUSDT counterpart, because the ETH perpetual's own history starts 2020-10-22. The first version of this measurement used one Bybit page and called it the full record, which it is not, and the long series below is the paginated one.
2. Converted every leg to dollars before dividing. Bybit reports open interest in coins, so each day's figure is multiplied by that same day's Binance daily close for the same asset, one price source for both legs so the ratio cannot inherit a price discrepancy. Binance publishes the dollar value itself and needs no conversion. OKX publishes dollars already.
3. Took the counting convention from Bybit's own API reference and then checked it against the data. The reference states that openInterest is the sum of both sides and singleOpenInterest the single side, in the base coin for a linear USDT contract. The data agrees: on all 400 rows pulled for the matched window the first is exactly twice the second, with a maximum residual of 0.01 that is pure rounding (2026-10-14: 56,177.505 against 28,088.753 BTC). The factor of two is identical on the BTC leg and the ETH leg, so it cancels out of a per-venue share whichever field you pick. It does not cancel out of the three-venue pool in the answer above, which is why that one figure is built from one-side values on all three legs rather than from the per-venue shares. Bybit is unusual in saying this plainly, which is the subject of a separate Athenum post cited at the end.
4. Matched contract scope where it was possible to match it, and said so where it was not. The Bybit and Binance legs are the same object on both venues: a single USDT-margined linear perpetual per asset. The OKX series is keyed by currency rather than by contract, so it sums that currency's contracts rather than one perpetual, and its level is therefore not strictly comparable to the other two. That is why the headline claim rests on the Bybit against Binance pair, which is scope-matched, and OKX is reported as a third reading rather than as the middle of a ranking.
5. Computed the share as ETH dollar open interest divided by the sum of the BTC and ETH dollar figures, per venue, per UTC day. No smoothing, no interpolation, no forward fill. Days where a venue did not publish both legs are dropped rather than filled.
6. Tested the share against the ETH/BTC price ratio from Binance daily closes, in levels, in first differences, and at leads and lags of one to three days, because the interesting version of the rotation claim is not that the two move together but that the leverage moves first.

The same question, three venues, one day

Prices are the last Binance daily-candle print at the 16:39 UTC pull, BTC 81,815.30 and ETH 2,438.79, with the 2026-10-14 candle still in flight rather than closed. Two stamping wrinkles, both measured rather than waved at. Dividing the Binance coin figure by its own dollar figure implies 83,125.70, which is exactly the 2026-10-13 close, so that daily row is a 00:00 UTC snapshot and not an end-of-day one. The same is true of the Bybit 1d row. Re-running the entire series against the price at each snapshot instant instead moves the headline from 27.99 to 28.15 percent, the matched-window median from 30.98 to 31.01 and the median venue gap from 11.93 to 11.90, leaving the gap positive on all 30 days and the two ranges still disjoint. The conclusions are insensitive to it; the third decimal is not.

The venue gap over 30 matched days, and the rotation over six years

The two matched-scope ranges, 27.99 to 33.77 and 40.34 to 43.89, share no value. A reader who wants to know whether leverage rotated into Ether over these 30 days gets a rising trend of +2.3 points from Binance and a falling trend of -2.7 points from Bybit, fitting a least-squares line to each. Endpoint to endpoint the same two series give +0.6 and -4.1 points, which is a reminder that the sign of a one-month "rotation" is partly a choice of estimator. Neither venue's answer is wrong about its own book.

A caveat that belongs with that table rather than after it: 30 consecutive daily readings are not 30 independent observations, and the claim above is deliberately not a p-value. Open interest is a stock, so today's share is mostly yesterday's share. What is being asserted is a persistent level difference of roughly 12 points between two venues measuring the same two contracts on the same days, which is visible without statistics, and specifically not that any one day's gap is significant.

For scale on the rotation itself, the largest single-day move in the Bybit share over the trailing 200 days is +4.50 points on 2026-07-15 and the largest fall is -4.18 points on 2026-06-25. Both are smaller than the standing gap between Bybit and Binance on an average day.

One question a reader should ask about any window that flatters a thesis is whether it was picked, so here is the test rather than the reassurance. Across all 1,985 overlapping 200-day windows in the full Bybit history, the trough-to-peak span of this share has a median of 15.28 points, a calmest window of 8.00 points (2023-05-12 to 2023-11-27) and a widest of 39.68 points (2020-10-28 to 2021-05-15). The trailing 200 days used above span 15.28 points and rank 1,051st of 1,985 from the calm end, and since 105 windows tie at that exact span, the window the API handed me is the median 200-day window rather than merely near it. So the window is typical of its length, and the thing that moves the answer is the length: pick 200 days and the venue gap matches the rotation, pick six years and the rotation is twice the gap. Per calendar year the span runs 29.21 points in 2021, 24.17 in 2022, 13.38 in 2023, 17.67 in 2024, 21.15 in 2025 and 15.28 so far in 2026.

The hypothesis the data killed

The version of the rotation claim worth money is a timing claim: leverage crowds into ETH before ETH/BTC moves, so the share is a leading indicator. Over the 200 Bybit days that is not supported.

In levels the share and the ETH/BTC ratio correlate at r = +0.638, which looks like a result and mostly is not: two series that each trend over six months will correlate in levels whether or not they are related. In daily first differences, which is the test that survives a trend, r = +0.305, contemporaneous and modest. The leads and lags are where the claim dies. With the share's daily change leading the price's daily change by one, two and three days, r is -0.021, -0.010 and -0.070 respectively, all indistinguishable from zero. The relationship is same-day and it does not point forward.

Sorting the 200 days into quartiles of the ETH/BTC ratio makes the magnitude plain as well: median share 28.41 percent in the cheapest quartile of ETH against BTC, then 28.96, 31.27 and 31.64 percent in the dearest. Across the entire 0.02578 to 0.03269 range of ETH/BTC the share shifts about 3.2 points, roughly a quarter of the gap between two exchanges on any given day.

The mechanism I can name, and the one I will not

The composition answer is available and it is ours. An Athenum post from 2026-07-12 broke open interest down per venue and found Binance holding 37.1 percent of Bitcoin perpetual open interest and 41.7 percent of the Ether pool, a venue whose share of the ETH market is larger than its share of the BTC market. That is the same fact as this one seen from the other side: if Binance is relatively more ETH-heavy than its peers, then Binance's ETH share of BTC plus ETH is mechanically higher than Bybit's, and today's three readings are three client mixes rather than three estimates of one quantity. The ratio is a property of a book, and it is reported as though it were a property of the market.

What I will not claim is why the mixes differ. Fee schedules, listing and margin tiers, the retail against institutional split, and the regional mix are all candidates and this measurement separates none of them. It also does not establish that the aggregate is the "true" share, because three venues are not the market, and a pool assembled from more venues would land somewhere else again. The finding is the dispersion and its persistence, not a correct value for the number.

Prior art, including our own

Our own nearest work comes first because it is nearer than anything external, and on this subject there are three pieces, not one. The 2026-07-12 open interest by exchange measurement cited above owns the per-venue concentration question, and reports each asset's pool separately rather than as a ratio. A second Athenum post from 2026-07-24 owns the denomination question, that the same open interest read in coins and in dollars can move in opposite directions, which is the trap step 2 above is designed around and the reason every leg here is converted before anything is divided. A third, from 2026-08-04, owns the counting convention outright: it establishes that most venues never say whether their open interest is one side or both, and that Bybit is the exception that documents it, which is the fact step 3 leans on rather than rediscovers. None of the three asks what this one asks: whether the ratio between two assets' leverage is stable across the venues that publish it. It is not, and the instability is the same size as 200 days of the rotation people quote the ratio for.

On the external side, the general result that venue-level order flow and positioning diverge in fragmented crypto markets is established, including in Albers, Cucuringu, Howison and Shestopaloff, "Fragmentation, Price Formation, and Cross-Impact in Bitcoin Markets" (arXiv:2108.09750, Applied Mathematical Finance 28(5), 2021, 395 to 448, published online 2022), with the caveat that it studies order books and trades for price formation rather than open interest, across a market set that is 11 perpetuals, 2 futures and 1 spot book. It motivates the measurement rather than supplying it. The operational consequence is the part I have not found stated with a number attached in the sources I checked: a cross-asset leverage ratio cannot be quoted without naming the venue it came from.

If you want to check any of this yourself, the per-exchange breakdown that the composition argument rests on is the thing to look at first. It is published, across the venues covered there and with Binance, Bybit and OKX among them, by Athenum.