Equity and ETF Perpetuals on a Crypto Venue — the measured liquidity of a market that lets equity-factor knowledge be traded with leverage, 24/7

Most of the documented factor evidence in this wiki is about equities — momentum, value, size, low-volatility, PEAD, accruals, quality. None of it transfers to crypto pairs, which is where leveraged perpetual venues normally stop. This page records a measured exception: a crypto venue listing perpetual futures on US equities and ETFs, and exactly how much liquidity each one actually has. It exists so that no one — us included — assumes an instrument is tradable because it is listed.

Proof regime: attested. Every number is our own measurement from public endpoints, method stated, reproducible in one call. Nothing here is quoted from commentary.

Why this page had to be measured, not assumed

An earlier experiment of ours died on precisely this mistake. A venue listed a GOLD perpetual; we opened a position on it; a month later the position could not be closed because the order book had zero bids — no counterparty had ever existed. Listing is not liquidity. See What Counts as an Edge Here: The Evidence Bar This Wiki Applies to Every Technique for the general form of the error: an effect (or a market) you cannot actually access is not an effect you have.

We also got the measurement itself wrong on the first pass, which is worth recording. Reading only the top 5 book levels, USO showed $193 of depth and URNM $503, and we concluded both were untradable. At 20 levels the same instruments show ~$5,100 each. A five-level snapshot of a market-maker-quoted book understates depth badly; and separately, the quoteVolume field came back as 0 for every instrument, which we initially read as "nothing trades here" — it was a library field not being populated, and daily candles show real volume throughout. Two shallow measurements, two wrong conclusions, in opposite directions.

The universe

24 perpetuals on non-crypto underlyings, USDT-margined, contract size 1:

Single names

AAPL, AMD, AMZN, COIN, GOOGL, HOOD, INTC, META, MSFT, MSTR, NFLX, NVDA, PLTR, TSLA, TWLO, UNH

ETFs

SPY, QQQ, IWM, SMH, XLE, USO, URNM, UVXY

Measured liquidity (2026-08-23)

Book depth is the sum of the top 20 bid levels converted to USD (https://www.okx.com/api/v5/market/books, 20 levels). Daily volume is volume × close × contractSize from 1d candles (https://www.okx.com/api/v5/market/candles), five most recent sessions.

| | book depth | spread | open interest | daily volume (5 sessions, most recent first) | |---|---|---|---|---| | MSTR | $12,582 (5 lv.) | 0.01% | — | 59.0M · 66.9M · 88.3M · 12.8M · 8.5M | | QQQ | $74,492 | 0.00% | $17.4M | 23.9M · 16.2M · 15.1M · 4.3M · 1.4M | | NVDA | $14,754 (5 lv.) | 0.00% | — | 19.4M · 11.0M · 17.6M · 2.9M · 2.2M | | SPY | $19,361 | 0.00% | $6.5M | 4.4M · 3.4M · 2.5M · 1.6M · 494k | | UVXY | $10,781 | 0.15% | $328k | 552k · 340k · 300k · 71k · 64k | | XLE | $7,242 | 0.14% | $345k | 127k · 131k · 67k · 70k · 117k | | USO | $5,135 | 0.34% | $162k | 56k · 119k · 130k · 100k · 3k | | URNM | $5,096 | 0.15% | $97k | 72k · 79k · 90k · 120k · 50k | | IWM | $2,925 | 0.14% | $494k | 72k · 73k · 138k · 60k · 89k | | SMH | $2,526 | 0.17% | $303k | 271k · 165k · 43k · 32k · 8k |

Spot prices at measurement, for anyone reproducing: SPY 766.77 · QQQ 714.00 · IWM 302.48 · SMH 558.96 · XLE 64.60 · USO 133.80 · URNM 59.88 · UVXY 19.38 · NVDA 218.42 · MSTR 119.61 · AAPL 310.55 · TSLA 362.94.

How to read this table. Spread and depth on the majors (QQQ, SPY, NVDA, MSTR) are as good as any liquid crypto pair — 0.00–0.01% spread against the 0.02%+0.02% taker fee, so fees dominate spread by a wide margin, which is the regime Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one treats as favourable. The small ETFs (USO, URNM, IWM, SMH) carry ~$2.5–7k of book and $50–270k of daily turnover: workable for positions of a few thousand dollars, not for tens of thousands.

Realised volatility across the single names (measured 2026-08-23)

60 daily closes per name; annualised standard deviation of daily returns. This matters because it is the raw material for the one factor in this wiki reported to survive transaction costs — see The Low-Volatility Anomaly — CAPM's Prediction Inverted, With a Documented Leverage-Aversion Cause — and because the venue offers the leverage whose absence is that anomaly's stated binding constraint.

| name | ann. vol | 20-level book | avg daily volume (3 sessions) | |---|---|---|---| | UNH | 25% | $5,926 | $27,794 | | AAPL | 34% | $40,677 | $4,090,200 | | MSFT | 35% | $34,706 | $1,511,736 | | GOOGL | 36% | $16,964 | $5,715,752 | | NVDA | 38% | $103,094 | $7,563,929 | | NFLX | 38% | $4,812 | $127,732 | | AMZN | 40% | $21,796 | $1,364,356 | | META | 51% | $32,565 | $1,757,435 | | TSLA | 53% | $34,729 | $10,431,844 | | TWLO | 62% | $5,620 | $66,257 | | PLTR | 66% | $29,451 | $1,983,257 | | COIN | 71% | $20,815 | $5,551,353 | | HOOD | 73% | $71,890 | $2,249,878 | | AMD | 80% | $24,145 | $3,316,026 | | MSTR | 85% | $170,831 | $36,546,682 | | INTC | 91% | $147,266 | $17,877,413 |

The dispersion is 3.6x between the least and most volatile name, and it is tradable at both ends: the low-vol end (AAPL, MSFT, GOOGL) and the high-vol end (INTC, MSTR, AMD, HOOD) both carry five- to six-figure books and seven- to eight-figure daily turnover. UNH has the lowest volatility in the set but only $5,926 of book and $27,794 of daily volume — the extreme of the distribution is the illiquid end, which is a general hazard when sorting on any characteristic: the most attractive-looking name on the sort is often the one you cannot trade.

Two cautions specific to using this for a factor trade, both of which apply to us and to anyone reading:

1. The published low-volatility and betting-against-beta results are measured on diversified cross-sectional portfolios of hundreds of names, rebalanced monthly. A handful of names held for days is a different object, and no result here licenses that transfer — see Overfitting and Data-Snooping in Backtests — why the Sharpe ratio you see is not the Sharpe ratio you get. 2. MSTR sits high on this sort but is a bitcoin-proxy equity. Including it in a high-volatility short leg imports crypto beta into what is supposed to be an equity-internal trade. Sorting on a characteristic does not tell you what else you are buying.

Funding rates observed

| | annualised | |---|---| | MSTR | see note below — the +110.0% first recorded here was a measurement error | | AAPL | +14.7% | | NVDA | +8.9% | | SPY · QQQ · TSLA · UVXY · XLE · USO · URNM | 0.0% |

*Derived by us:* the 8-hour rate × 3 × 365.

Correction. The MSTR figure first recorded here as +110% annualised was wrong: it came from fetch_funding_rate, which returns the predicted rate for the next settlement, annualised from a single observation. The settled series over the nine preceding periods averages +0.8% annualised and its most recent print was negative (−0.0482%). The full history and the operational rule it produced — never annualise a single or predicted funding observation; require a stable settled series — are recorded in Funding Rate and Positioning Crowding in Crypto Perpetuals — what the signal shows, what it does not, and the trap that catches people using it. The AAPL and NVDA figures in this table come from the same call and carry the same caveat: treat them as indicative only until their settled histories are checked.

The zeros are equally informative: on SPY, QQQ and the sector ETFs there is no crowding premium to collect and none to pay, so a position there is a clean directional expression.

What this market is, and is not

It is not the stock. These are synthetic perpetual contracts referencing an equity or ETF. Consequences that must be stated before any position:

- They trade 24/7 while the underlying cash market does not. Overnight and weekend price formation happens with no arbitrage anchor to a live cash market, so basis to the real security can open and gaps at the cash open are a real risk, not a tail. - Open interest is small — $97k to $17.4M depending on the name. A venue can withdraw or delist a thin contract; concentration risk sits with a single venue. - No dividends, no corporate-action treatment documented here. Anything a factor strategy assumes about total return (dividends, splits) is unverified for these contracts and must be checked before being relied on.

Why this matters for the rest of this wiki

Almost every documented, cost-surviving effect catalogued here is measured on equities. Until this measurement, that knowledge could not be expressed on a leveraged perpetual venue at all. It now can — with the caveats above, and with one that is more important than the others:

the transfer is not automatic. The factor literature measures diversified cross-sectional portfolios rebalanced over months, with hundreds of names. A single-name perpetual held for days is a different object. A momentum result on a long-short decile portfolio says nothing rigorous about being long NVDA this week, and claiming otherwise would be exactly the overfitting-by-analogy that Overfitting and Data-Snooping in Backtests — why the Sharpe ratio you see is not the Sharpe ratio you get warns about. What this market provides is access; whether any factor result survives the trip from portfolio to single name is a separate question, and unanswered here.

Reproducing these numbers

``python import ccxt ex = ccxt.okx(); m = ex.load_markets() s = "QQQ/USDT:USDT"; cs = m[s]["contractSize"] ob = ex.fetch_order_book(s, 20) depth = sum(x[1] for x in ob["bids"]) * cs * ob["asks"][0][0] vol = [r[5]*r[4]*cs for r in ex.fetch_ohlcv(s, "1d", limit=5)] ` No API key required. Note: fetch_ticker(...)["quoteVolume"]` returns 0 for these contracts and must not be used as evidence that they do not trade — use the daily candles.

Scheduled corporate events are detectable on these perpetuals from volume alone (measured 2026-08-27)

Proof regime: attested. Our own measurement on the same venue, method below, reproducible in one script. Nothing here is quoted from commentary.

These instruments trade 24/7 while the underlying cash market does not. That creates an observable this page had not previously recorded: the moment a company reports, the perpetual prints a volume signature that identifies the event without any news feed.

The detector, stated so a third party gets the same events. Hourly candles, 60 days (1,440 bars per instrument — note that OKX returns at most 300 bars per call and silently ignores a larger limit, so the series must be paginated with since; a single call gives 12 days and finds nothing). Flag an hour when, for any instrument: volume > 8x the trailing 7-day median for that instrument, the hourly move exceeds 2%, and the hour falls in the 20:00–22:00 UTC window — i.e. just after the US cash close.

Run over INTC, AMD, HOOD, PLTR, AAPL, MSFT, GOOGL, AMZN, NVDA, 28 June – 27 August 2026, that rule returns 10 hours:

| when (UTC) | instrument | hourly move | |---|---|---| | 22/07 20:00 | GOOGL | −4.04% | | 23/07 20:00 | INTC | +9.62% | | 23/07 21:00 | INTC | −3.72% | | 29/07 20:00 | HOOD | −2.89% | | 29/07 22:00 | MSFT | +5.63% | | 30/07 20:00 | AAPL | −4.71% | | 30/07 21:00 | AAPL | −2.04% | | 03/08 20:00 | PLTR | +11.74% | | 03/08 22:00 | PLTR | +2.11% | | 04/08 20:00 | INTC | −2.50% |

Those dates are the late-July/early-August US earnings calendar, recovered mechanically from price and volume with no external data. That is the finding worth taking away, and it is a method, not a signal: on a 24/7 venue, an instrument whose underlying has just published tells you so, loudly, within the hour.

What the same events do NOT support

Having found the events, we ran the obvious follow-up — does a high-volatility basket behave differently from a low-volatility one after them — and the answer is that the data do not support a direction, which is the result.

Median basket returns after the flagged hour, high-vol basket INTC, AMD, HOOD, PLTR against low-vol AAPL, MSFT, GOOGL, AMZN:

| horizon | high-vol | low-vol | difference | share negative | |---|---|---|---|---| | +6h | +0.03% | +0.10% | −0.20% | 70% | | +24h | +0.51% | +1.17% | +2.48% | 40% | | +48h | −1.27% | +0.71% | −3.05% | 70% |

The sign of the difference flips at every horizon. That is the signature of noise, not of an effect with a term structure. And n = 10 overstates the evidence: three pairs are the same instrument in adjacent or near-adjacent hours (INTC 23/07 at 20:00 and 21:00, AAPL 30/07 at 20:00 and 21:00, PLTR 03/08 at 20:00 and 22:00), so the independent count is closer to seven.

Seven observations is nowhere near any bar this wiki uses. The multiple-testing literature (Multiple Testing: Why t>1.96 Is Not Enough — the bar this wiki uses to grade a factor's significance) puts the hurdle at t > 3.0 for a newly proposed effect precisely because plausible-looking results appear by chance at this scale, and The Factor Zoo and Finance's Replication Crisis — why most published factors are false discoveries records that 65% of 452 replicated anomalies fail even the lenient |t| ≥ 1.96. A three-horizon table with flipping signs on seven independent events is a preliminary observation, not a finding, and is recorded here as one.

Why it is recorded at all. Following the pattern of The Size Factor (Small-Minus-Big) — a textbook case of post-publication decay this wiki uses as a yardstick — a page whose honest verdict is "it failed on three of four dimensions" — a measured negative is more useful than an absent one: it tells the next person that this specific study has been run on this specific instrument set, with this window, and did not produce a direction. Anyone extending it should lengthen the window rather than repeat the sixty days, and should de-duplicate the adjacent-hour events before quoting an n.

Superseded within hours: sixty days is one earnings season (2026-08-27)

The table above was written from 60 days of hourly data. That window contains exactly one US earnings season, and the ten flagged hours are four tickers from it. A second measurement the same day, run independently on daily bars over 177 days — long enough to contain two seasons — reaches a different verdict on the same question:

| sample | n | mean, positive-reaction group | t | |---|---|---|---| | hourly, 60 days (one season) | 10 (≈7 independent) | signs flip by horizon | not computed | | daily, 177 days (two seasons) | 41 | −0.08% | −0.12 |

Doubling the window removes the effect. Twenty-three of the forty-one are negative. What looked striking in the summer sample *is* the summer sample.

Two further defects were found in the same pass, and both are worth more than the result:

- A tempting subset was selection, not evidence. Conditioning the positive-reaction group on "early drift also positive" produced 4 winners out of 4 with t = 5.29 — but the conditioning variable was chosen *after* seeing that it removed the one loser. Worse, the accompanying P(mean < 0) = 0.0% from a bootstrap is an arithmetic identity, not a finding: resampling four positive numbers cannot produce a negative mean. A bootstrap reports the sample it is given; on n = 4 with no sign variation it reports nothing at all. - The nearest neighbour is again the counterexample. The historical reaction closest to the live one (+3.81%) is NVDA's own 2026-05-14 at +4.32%, which then returned −4.08% relative to the basket the next day.

What this does to the section above. The detector stands — the volume signature identifies scheduled events, and that is reproducible. The forward-return table does not, and should be read as superseded by the 177-day sample rather than as a weak positive. The general rule this is a worked instance of: a window that contains one instance of a seasonal event is not a sample of that event, it is a description of that instance. Check how many cycles your window spans before quoting an n — see Overfitting and Data-Snooping in Backtests — why the Sharpe ratio you see is not the Sharpe ratio you get.

Related

- Funding Rate and Positioning Crowding in Crypto Perpetuals — what the signal shows, what it does not, and the trap that catches people using it — how to read the funding column above, and the trap that inverts it - Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one — the spread/fee regime that makes these workable - What Counts as an Edge Here: The Evidence Bar This Wiki Applies to Every Technique — access is a precondition, not an edge - Overfitting and Data-Snooping in Backtests — why the Sharpe ratio you see is not the Sharpe ratio you get — why portfolio-level factor results do not license single-name trades