Historical decision study · By CoinNudge Research · · Guide updated 2026-10-04 · 15 min read
Extreme Funding + Rising OI: Crowding Is Not a Forecast
Extreme funding plus rising open interest is a crowding condition, not a guaranteed reversal. This short Hyperliquid pilot retained only four positive-extreme/rising-OI cases; all four subsequently fell over 24 hours, but they span just two UTC days. Negative-extreme/rising-OI candidates were all excluded by the predefined cooldown. There is no credible general squeeze probability yet. The value here is the auditable event ledger, price paths and a method that does not confuse rising dollar OI with new positions.
Preview alert list · Explore research data · See alert plans
Key points
- 392 retained observations; see complete/pending denominators for each outcome.
- Hourly funding quotes; quantity OI; 15-minute mark-price observations.
- Short-history exploratory pilot. Not settled funding, execution PnL or a universal squeeze probability.
What happened after extreme funding and rising OI—in the actual sample?
Only four positive-extreme/rising-quantity-OI cases survived the Hyperliquid cooldown; all four had negative subsequent 24-hour mark changes—not demonstrated profitable short trades. The frozen archive contains 149,388 public snapshots across 177 contracts, from 2026-09-09T05:00:00Z to 2026-09-18T00:00:00Z (end exclusive). After input checks and the shared 24-hour cooldown, 392 observations remain across five predefined groups. The large input-row count is not the number of independent extreme events.
The positive-extreme/rising-OI group contains CHIP, PURR, USELESS and VVV. Their 24-hour reference-price changes were negative, but three occurred on the same UTC date and the fourth on the next. Reporting this as a “100% reliable short signal” would be misleading. The non-extreme/rising-OI comparison also had a negative median, indicating broader market weakness rather than an isolated mechanism.
This edition answers how the condition should be measured and what this small recorded sample did. It cannot establish whether negative funding reliably precedes a short squeeze, whether a trade is profitable after funding, or whether the relationship generalizes across venues and regimes.
| Group | Retained | 4h complete | 4h median mark change | 24h complete | 24h median mark change | Distinct 24h event dates |
|---|---|---|---|---|---|---|
| Positive extreme + rising OI | 4 | 4 | -1.42% | 4 | -10.63% | 2 |
| Positive extreme; OI below +2% | 29 | 28 | -0.59% | 25 | -1.69% | 8 |
| Negative extreme + rising OI | 0 | 0 | — | 0 | — | 0 |
| Negative extreme; OI below +2% | 30 | 30 | +0.01% | 27 | -0.78% | 8 |
| Non-extreme funding + rising OI | 329 | 324 | -0.13% | 279 | -1.46% | 8 |
What counts as extreme funding, rising open interest and a usable outcome?
Extreme means a current Hyperliquid quote at least +0.01% per hour or at most −0.01% per hour. Rising means underlying quantity OI increased by at least 2% over one hour; the current notional OI must be at least $1 million. We require all five consecutive fifteen-minute snapshots across that hour.
Quantity OI is recovered as oi_usd / mark_price from the same record. The collector originally formed notional OI as quantity × mark, so dividing by that mark reverses the conversion. A 10% price rise with unchanged positions makes dollar OI rise 10% but quantity OI rise 0%; it is not new position growth.
The reference is the next fifteen-minute snapshot. Outcomes are its mark-price change after four or twenty-four hours. Every intermediate snapshot must exist; missing and not-yet-mature windows stay null. MFE and MAE are extrema of sampled mark changes, not intrabar price extrema or leveraged profit and loss.
How do you tell whether a funding rate is really extreme for that coin?
Match the venue, contract and time unit before judging an extreme funding rate. This pilot uses a current Hyperliquid quote of at least ±0.01% per hour; it is not a threshold for every exchange or a measured settlement total.
A separate historical-normality test would compare the same contract’s current reading with only its preceding valid observations, using a declared window and percentile method. It must report sample size, gaps and the cutoff. Different settlement intervals need equivalent time units but simple normalization cannot remove venue or contract differences.
A percentile calculated from the whole finished archive would use future observations to classify an old event. Do not call that time-available. The published pilot uses its fixed threshold, not a newly validated rolling-percentile classifier; small or missing historical samples should stay unclassified rather than acquire an invented percentile.
| Measure | What it answers | Boundary |
|---|---|---|
| Current hourly quote | Cost/crowding indicated now on this venue | Not the final amount settled |
| Historical percentile | Unusual relative to preceding same-contract data | Needs a frozen trailing sample and valid history |
| Settled funding | Amounts and times actually paid/received | Requires settlements and position exposure |
| Quantity OI change | Whether open quantity expanded | Does not reveal every trader’s direction |
Does one extreme quote mean funding will stay extreme?
No. A single extreme quote proves neither a sustained episode nor the final funding payment. In this pilot, five consecutive qualifying snapshots establish the study’s local persistence condition; they do not establish that the quote continued for hours or through settlement.
A persistence study needs a complete ordered sequence, an allowed maximum gap, episode start/end rules and right-censoring when the source or archive ends. Missing quotes must not be forward-filled into a continuous funding run. Quote persistence and consecutive extreme realized settlements should be separate outputs.
The four retained positive events occurred on only two UTC days, and all retained negative-extreme/rising-OI candidates were lost to the shared cooldown. This cannot estimate reliable sign-specific persistence or squeeze odds. Wider archived funding and OI data would support a separate, versioned study, not a reinterpretation of these four records.
Verify: Historical funding and OI coverage, gaps and delivery
Can funding income offset a losing price move—or protect you from liquidation?
Funding income does not eliminate directional price loss or liquidation risk. Keep position notional, rate interval, holding duration, mark-price P&L, fill costs and margin separate. The pilot’s forward returns are mark-price changes before funding and costs, not a net trade ledger.
As an explicitly hypothetical example, an unchanged $1,000 short receiving +0.05% every eight hours would collect $1.50 over three daily settlements. If that same quoted percentage applied every hour, 24 settlements would total $12.00. Those are different contracts; neither can be substituted for the other. A 2% adverse mark move loses roughly $20 before costs in the simple fixed-quantity example, exceeding either funding amount.
With $100 margin and $1,000 notional, that $20 move consumes about 20% of initial margin before other effects. Real liquidation depends on maintenance margin, venue rules, collateral and fees. This arithmetic does not model a venue liquidation price. A hedged spot/perpetual strategy also needs both legs, basis and capital, not just a favorable funding sign.
| Illustrative fixed $1,000 notional | One day amount | Meaning |
|---|---|---|
| +0.05% per 8h, three settled payments | +$1.50 to the short | Only if position and the assumed rate persist |
| +0.05% per 1h, 24 settled payments | +$12.00 to the short | Different settlement cadence |
| 2% adverse move for a fixed-quantity short | Approximately −$20.00 | Price risk before fill costs and other charges |
| Published pilot’s next-24h outcome | Unlevered mark change | Not any of these funding-inclusive net amounts |
What should you check before trading an extreme funding and rising-OI signal?
First verify contract identity, source times, funding unit and whether the figure is a quote or a settlement. Then separate quantity OI expansion from a dollar-OI increase caused by price. A valid crowding observation is still not permission to fade a strong trend.
Compare spot participation, price structure, liquidity and the distance to your invalidation level. Do not use the future 24-hour result as evidence that was available at entry. Estimate both directional loss and full holding costs; if a hedge is proposed, reconcile both legs and their capital requirements.
Wait when information is stale, the period is unknown, execution cannot be sized, or your only rationale is “four previous cases fell.” A useful action may be monitoring a crowded condition rather than opening a leveraged position. The current sample cannot justify a universal long/short recommendation.
Verify: Compare current funding rates, units and carrying costs · Inspect a fee-aware funding-arbitrage experiment · Check spot demand versus leverage
Does positive funding mean longs are crowded? Does negative funding mean a squeeze?
At a positive funding payment, longs pay shorts; at a negative payment, shorts pay longs. This page samples a current quote, not the final settlement record. The quote can change before payment. Multiplying an hourly quote by eight is a constant-rate illustration, not an observed eight-hour funding cost.
Open interest counts outstanding contracts. Every contract has two sides, so increasing OI alone does not tell you that there are “more longs than shorts.” Funding, price direction, spot buying and observed liquidation context can help frame competing explanations, but they do not identify the eventual forced side by themselves.
A rising price with growing OI and positive funding can reflect leveraged demand, hedging or multiple offsetting participants. A falling price with negative funding can continue falling rather than squeeze. To call something a liquidation cascade requires liquidation evidence; these three fields alone do not provide it.
| Observed combination | Question to investigate—not a trade instruction |
|---|---|
| Positive extreme + OI growth + rising price | Is spot demand confirming the rise, or is the move increasingly dependent on leverage? |
| Positive extreme + OI growth + falling price | Are positions adding into weakness? Look for actual reductions/liquidations rather than presuming them. |
| Negative extreme + OI growth | Is the price resisting selling? Negative carry by itself does not prove a short squeeze is imminent. |
| Extreme funding + flat/falling quantity OI | Is crowding already unwinding? A dollar-OI move alone cannot answer. |
Verify: Hyperliquid funding mechanism
Why are there only four positive cases—and zero retained negative cases?
We keep the earliest qualifying candidate per contract, then exclude new candidates from that contract for twenty-four hours, regardless of group. This reduces repeated observations of the same episode. It also creates a material selection limitation: an ordinary funding/OI-growth observation can occupy the cooldown before a later extreme arrives.
There were 34 raw negative-extreme/rising-OI qualifying snapshots, but every one fell inside that shared cooldown. Zero retained cases therefore does not mean no negative-extreme condition occurred. Similarly, four retained positive cases came from 44 qualifying snapshots. Full excluded rows are downloadable, making this limitation visible rather than hiding it.
This is a prespecified first-opportunity study, not a complete census of threshold-crossing episodes. A future separately versioned study could examine transitions into extremes with a different episode definition; it must not silently replace this cohort or claim an independently validated improvement. Groups here are unmatched for symbol, hour or volatility, so their median differences are descriptive.
| Group | All qualifying candidates | Cooldown excluded | Retained | 24h pending | 24h missing |
|---|---|---|---|---|---|
| Positive extreme + rising OI | 44 | 40 | 4 | 0 | 0 |
| Positive extreme; OI below +2% | 293 | 264 | 29 | 4 | 0 |
| Negative extreme + rising OI | 34 | 34 | 0 | 0 | 0 |
| Negative extreme; OI below +2% | 685 | 655 | 30 | 3 | 0 |
| Non-extreme funding + rising OI | 2541 | 2212 | 329 | 50 | 0 |
All retained positive-extreme/rising-OI observations
The retained group is CHIP, PURR, USELESS and VVV on Hyperliquid, over only two UTC dates, with the next fifteen-minute mark as the reference. All four are shown, not a hand-picked winner gallery. The final column is an unlevered reference-price change; it is not a short trade return, funding-income estimate or executable profit. During the path, every case also experienced some upside from the next-snapshot reference. A negative final price change does not mean a short position had no adverse excursion.
These contracts are not BTC or ETH. Small-asset conditions and the short market window cannot establish a Bitcoin-specific rule. The BTC/ETH records remain in the input directory even though they did not populate this extreme subgroup under these rules.
| Contract | Decision boundary UTC | Quote %/hour | Quantity OI 1h | 24h mark change | Sampled upside | Sampled downside |
|---|---|---|---|---|---|---|
| CHIP-PERP | 2026-09-09T07:15:00Z | +0.01040% | +5.11% | -14.68% | +5.79% | -14.80% |
| PURR-PERP | 2026-09-09T07:30:00Z | +0.01729% | +2.44% | -6.57% | +1.82% | -8.20% |
| USELESS-PERP | 2026-09-09T06:15:00Z | +0.01003% | +7.88% | -29.44% | +1.96% | -32.72% |
| VVV-PERP | 2026-09-10T17:15:00Z | +0.01330% | +2.10% | -1.56% | +8.18% | -5.89% |
Historical path: CHIP · 2026-09-09
This Hyperliquid CHIP · 2026-09-09 case uses the next fifteen-minute mark as reference, not an executed order. The snapshot quoted +0.01040% hourly funding with +5.11% quantity-OI growth over one hour. The subsequent 24h mark change was -14.68%; sampled upside reached +5.79% and sampled downside -14.80% relative to the next-snapshot reference.
Top: mark price. Middle: current hourly funding quote, which may change before payment. Bottom: quantity OI, not dollar OI. S = signal snapshot; R = next-snapshot reference; X = 24h endpoint. The markers are approximate snapshot boundaries, not guaranteed execution times. All retained positive+rising examples are displayed, including their interim upward moves; none is a leveraged trading backtest.
Historical path: PURR · 2026-09-09
This Hyperliquid PURR · 2026-09-09 case uses the next fifteen-minute mark as reference, not an executed order. The snapshot quoted +0.01729% hourly funding with +2.44% quantity-OI growth over one hour. The subsequent 24h mark change was -6.57%; sampled upside reached +1.82% and sampled downside -8.20% relative to the next-snapshot reference.
Top: mark price. Middle: current hourly funding quote, which may change before payment. Bottom: quantity OI, not dollar OI. S = signal snapshot; R = next-snapshot reference; X = 24h endpoint. The markers are approximate snapshot boundaries, not guaranteed execution times. All retained positive+rising examples are displayed, including their interim upward moves; none is a leveraged trading backtest.
Historical path: USELESS · 2026-09-09
This Hyperliquid USELESS · 2026-09-09 case uses the next fifteen-minute mark as reference, not an executed order. The snapshot quoted +0.01003% hourly funding with +7.88% quantity-OI growth over one hour. The subsequent 24h mark change was -29.44%; sampled upside reached +1.96% and sampled downside -32.72% relative to the next-snapshot reference.
Top: mark price. Middle: current hourly funding quote, which may change before payment. Bottom: quantity OI, not dollar OI. S = signal snapshot; R = next-snapshot reference; X = 24h endpoint. The markers are approximate snapshot boundaries, not guaranteed execution times. All retained positive+rising examples are displayed, including their interim upward moves; none is a leveraged trading backtest.
Historical path: VVV · 2026-09-10
This Hyperliquid VVV · 2026-09-10 case uses the next fifteen-minute mark as reference, not an executed order. The snapshot quoted +0.01330% hourly funding with +2.10% quantity-OI growth over one hour. The subsequent 24h mark change was -1.56%; sampled upside reached +8.18% and sampled downside -5.89% relative to the next-snapshot reference.
Top: mark price. Middle: current hourly funding quote, which may change before payment. Bottom: quantity OI, not dollar OI. S = signal snapshot; R = next-snapshot reference; X = 24h endpoint. The markers are approximate snapshot boundaries, not guaranteed execution times. All retained positive+rising examples are displayed, including their interim upward moves; none is a leveraged trading backtest.
Can this be used to backtest a funding-and-OI strategy?
This Hyperliquid funding-and-OI pilot can help validate event definitions and data availability, but cannot estimate trade success or justify position size. Its less-than-nine-day archive retains four positive-extreme cases over two UTC dates and no negative-extreme cases. Correlated assets and unmatched groups prevent an out-of-sample performance claim; no statistical interval is justified for a four-case probability.
A serious strategy study needs a longer multi-regime archive, final funding settlements, executable bid/ask or order-book assumptions, a frozen historical universe, known-at timestamps and explicit risk/margin rules. Retrospective mark paths cannot supply those missing ingredients. Snapshot MFE/MAE also misses excursions between samples.
For a trader, the useful workflow is to inspect funding units, verify quantity OI, check whether price and spot demand agree, then revisit the actual next observation. For a researcher, this release provides explicit units, the full first-opportunity ledger, pending outcomes and the negative-case exclusion audit. Those are verifiable contributions even though the short sample cannot yet support a reliable trading edge.
Reproduce the experiment: inputs, candidate ledger and field definitions
Download candles.json, derivatives.json and three_decision_studies.py into the same directory. Run python three_decision_studies.py . reproduced.json. The standard-library-only program rebuilds all three studies. Compare the result with research.json. The JSON keeps full precision; CSV nested objects are JSON-encoded, not silently truncated. Chart coordinates and selection rules are in chart-data.json.
The SHA-256 manifest verifies each published artifact. These are frozen September 18, 2026 results, not a live ticker. An update must publish a separate evidence edition and revise the answer, tables, FAQ and Dataset together. These public study files are free to download. Separately purchased fixed-edition data packs and optional API access have their own listed contents and permissions; this experiment does not expand those rights.
The derivative archive retains event_time as the bucket start and archived_observed_at as its archive observation time. Those are not interchangeable. Rows were collected locally, but historical archive timestamps do not prove the first moment every value was available to a trader. A conservative decision boundary is bucket start + 15 minutes; the next bucket supplies the reference mark. Even this is a snapshot study, not a latency-correct execution simulation.
| Field / artifact | Definition and missing-value rule |
|---|---|
| return | Fractional reference-price or net account change. 0 means a real unchanged outcome or a defined cash policy; missing remains null. |
| status | complete, pending, missing; dip/funding candidate ledgers also retain cooldown exclusions. |
| event_time / UTC | DCA allocation time; dip closed-candle decision boundary; funding historical snapshot bucket start. See decision_boundary for funding. |
| policies / outcomes | Spot: entered flag, execution times/prices and costs. Funding: 4h/24h reference times/prices and sampled excursions. |
| taker_buy_valid | Only valid=1 with positive quote volume can define an active-buy ratio. A missing ratio is not 100% selling. |
| observations / candidates | Retained observations are not every snapshot. The candidate ledger exposes missing, pending and cooling-down candidates. |
| uncertainty | Dip paired mean differences have clustered intervals. DCA overlap and the very short derivative pilot are described explicitly, not assigned spurious precision. |
Verify: Full analysis JSON · This study: complete candidate CSV · This study: summary CSV · Candle inputs JSON · Candle inputs CSV · Derivative input JSON · Derivative inputs CSV · Standalone reproduction program · Chart coordinates · Preset protocol · Integrity manifest
Take the next step with this evidence.
Choose ongoing Telegram monitoring or inspect structured data in your own research workflow.
Receive qualifying events in Telegram.
Use the ready-made alert checklist, message explanations and daily briefing. Check the supported categories and coins before subscribing; every Research page is not a separate alert category.
Check the alert list · View alert plans
Inspect the data before choosing a pack.
Prepared research packs supply scoped features, separate outcomes where available and quality flags. Buy a fixed edition once and download directly; API use is optional.
Check data pack samples · Explore the full data catalog · Compare data plans
Each purchased pack includes its specified fields and published date range. This page's full dataset is not automatically included. Other data families require coverage, date, field and source-use confirmation before purchase. Telegram Pro is a separate 30-day service; historical packs are one-time purchases.
Frequently asked questions
What happens after extreme funding and rising open interest?
In this short Hyperliquid pilot, four retained positive-extreme/rising-OI cases fell over the next 24 hours. They span only two dates and are not sufficient to infer a general reversal probability.
Does high positive funding mean I should short?
No. A positive quote is a funding/crowding observation, not proof that price must fall. The quote may change and a short can experience adverse moves or liquidation even if the later mark is lower.
Does negative funding predict a short squeeze?
This edition cannot estimate that. Thirty-four negative-extreme/rising-OI snapshots qualified before deduplication, but all were excluded by the shared 24-hour cooldown. Zero retained does not mean zero occurrence.
Why use quantity OI instead of dollar open interest?
Dollar OI rises when the underlying price rises even if positions are unchanged. Dividing the collector’s notional OI by its same-record mark recovers the quantity used in this study.
What funding rate is considered extreme here?
An absolute current quote of at least 0.01% per hour on Hyperliquid. It is a declared study threshold, not a universal threshold across exchanges with different settlement intervals.
Are the 24-hour results net of funding and trading costs?
No. They are unlevered mark-price changes. Final funding settlements, fill prices, fees, margin and liquidation mechanics are not modelled.
Can I apply these cases to BTC and ETH?
Not as a measured BTC/ETH result. The four retained positive-extreme cases are CHIP, PURR, USELESS and VVV; this small-asset subgroup does not establish a Bitcoin or Ethereum rule.
Is the archive strictly point-in-time?
Not proven by these fields. The bucket start and archive observed time are preserved separately; neither automatically proves a trader’s first access time. The article explicitly labels that temporal limitation.
Can I download excluded and pending observations?
Yes. The candidate CSV retains cooldown exclusions and pending/missing outcomes; the raw input directory and reproduction program are public.
Does one high funding reading mean it stays high all day?
No. The pilot’s five-snapshot condition is local persistence, not a full-day settled rate. A run study needs episode and gap rules; final funding settlements are separate.
Can I multiply an eight-hour funding rate by 24?
Not to estimate one day at that cadence. With an unchanged assumed rate there are three eight-hour payments in a day; 24 payments describe hourly funding. Actual rates, exposure and settlement records can change.
Sources and verification
- Binance spot candlestick definitionsOfficial OHLC, quote-volume and taker-buy field definitions.
- Investor.gov: dollar-cost averagingDefinition of regular equal-sized investing, not evidence for this BTC result.
- Hyperliquid funding documentationHourly funding mechanics and payer direction; quotes are not guaranteed future settlements.
- CoinNudge experiment and evidenceFull-precision observations, denominators, groups and sensitivity.
Inspect the service before choosing a subscription.
See supported Telegram alerts and research-data samples. These experiments do not execute trades.
Explore alert coverage