BVIV (Volmex Bitcoin Implied Volatility Index)

BVIV is the Volmex Bitcoin Implied Volatility Index: the market's expected volatility of bitcoin over the next 30 days, derived from bitcoin option prices and quoted in volatility points.

The BVIV perpetual lets you take a view on bitcoin volatility itself rather than on the bitcoin price. Buying the perp takes a long view on 30-day implied volatility; selling takes the opposite view. What you make or lose also depends on the price you trade at, on funding and on fees.

There is no BVIV spot. You cannot hold the index, and there is no underlying asset to deliver or arbitrage against. Price discovery depends on the published index and on market makers quoting around it.

Before you trade

• Funding follows a model of the volatility curve. When the curve slopes upward, longs pay a carry that ran at the equivalent of about 250% to 350% a year on 25 and 26 September 2026. When the curve flattens or inverts, the sign reverses and shorts pay.
• The funding clamp starts narrow. It is 3 basis points per 8 hours at launch and will be raised gradually to 30, with notice. While the clamp is narrower than the modelled carry, most of the carry is paid only if the market trades at a premium to the index.
• The mark price follows the order book. In testing, one participant holding the book 12% below an unchanged index for about seven minutes moved the mark 11% and liquidated maximum-leverage positions. The mark’s basis is currently not capped, so only Hyperliquid's 20% order price limit stops it.
• If the index stops updating, the oracle can drift toward this market's own order book, within bounds. See Degraded mode and the internal oracle.
• Isolated margin only, up to 5×. Liquidations go to the order book. No backstop liquidator covers this market, so if a position's value falls below zero before the book absorbs it, auto-deleveraging closes positions on the other side.

Read Risks and disclosures before trading.

Contract specification

Isolated only. BVIV margin is held apart from your other positions. A BVIV liquidation cannot take collateral from them, and theirs cannot take BVIV margin. The market is strict isolated: you can add margin to an open position but cannot withdraw it. Margin is released in proportion as you reduce the position.

Leverage and liquidation distance. You can choose 1× to 5×. Initial margin is position value divided by your leverage. Maintenance margin is 10% of position value, half the initial margin at 5×. Measured on the mark price from your entry, before funding, fees and any margin you add, you are liquidated after these moves:

Funding payments come out of your margin and move these distances closer.

The BVIV index

BVIV is built and published by Volmex. Its methodology paper⁠ (July 2026) sets out the full construction. In outline:

• Sources. Bitcoin option and futures quotes from Deribit and OKX, merged into one global option book. The underlying price is taken from the venue with the most open interest, currently Deribit.
• Method. Model-free variance-swap replication. Out-of-the-money calls and puts at the two expiries either side of 30 days give two implied variances, which are interpolated to a constant 30-day maturity.
• Smoothing. Volmex smooths the raw variance with an exponentially weighted moving average and then takes the square root: BVIV = 100 × √(smoothed variance). The half-life is one minute, and two minutes between 07:30 and 08:30 UTC, around the daily option expiry.
• Publication. Every second, every day. Volmex also publishes 1, 7, 14, 60, 90, 120 and 180-day tenors. Funding on this market uses the 7, 14, 30 and 60-day prints.
• Exceptions. When the replication cannot be computed, or comes out below at-the-money Black–Scholes implied volatility, Volmex substitutes an estimate built from at-the-money implied volatility and a smoothed measure of how richly the tails are priced.

Two points on the index:

The index is smoothed. Smoothing damps single-quote noise in the option book, but the index lags a sudden repricing of options by about one half-life. It is not a raw tick print.

The index is independent of this venue. BVIV is computed from the bitcoin options market, not from trading on mkts. Order flow and positioning here do not feed back into the index. They do move the mark price and the funding premium, and while the index is stale they can move the oracle. See Degraded mode and the internal oracle.

Oracle design

The oracle price for BVIV is the latest valid Volmex BVIV print and relayed on-chain by Kinetiq roughly every three to four seconds.

The oracle runs two programs for this market: a price feed and a funding feed. Kinetiq, as deployer of record, calls both and relays their outputs on-chain with setOracle and setFundingInterestRates. It does not compute either value.

Data flow:

Nothing downstream of the relayer feeds back into the index. The price feed does read this market's order book: always for the mark basis, and for the oracle itself only while the index print is stale.

What each price update contains. Every update carries three prices, and Hyperliquid uses each one differently.

Price increments. Published prices have at most five significant figures and at most four decimal places, which is six minus the market's two size decimals. Whole numbers are exempt from the significant-figure rule. The oracle rounds to significant figures first, then to decimal places.

Protocol rules. Hyperliquid applies its own rules to every update:

• At least 2.5 seconds must pass between updates. Kinetiq aims for one about every three to four seconds.
• The mark price can move at most 1% from the previous mark in one update.
• All prices are clamped to 10× their start-of-day value.
• If no update arrives for 10 seconds, the mark price falls back to the local book mark alone.

Mark price

The mark price is used for margin, unrealised P&L, liquidations, and take-profit and stop-loss triggers. Funding payments use the oracle price, not the mark.

L is Hyperliquid's own local mark. The mark is then moved at most 1% from the previous mark.

basis, updated by the oracle on every price update:

• mid = (best bid + best ask) / 2 on mkts:BVIV
• dt = seconds since the previous update
• basis = 0 while the book is not two-sided
• basis is not capped — it can reach the full gap between the book and the oracle

M2 uses the same construction Hyperliquid uses for its own perps: the oracle plus an exponential moving average of the book's premium, with a 150-second time constant.

How this behaves:

• When the book agrees with the index, the three inputs sit together and the mark is the oracle.
• When the book moves away and stays there, the mark follows M2 with a lag.
• When the book comes back, the mark returns almost at once, because the local mark and the oracle agree again and outvotes M2.
• When either side of the book is empty, oracle sets the basis to zero, M1 and M2 are both the oracle, and the mark returns to the oracle at up to 1% per update.
• The basis is not capped. Oracle does not limit how far M2 can move from the oracle. A book held away from the index long enough carries the mark with it, as far as Hyperliquid's order price limit, 20% at 5×. The median described on Mark price stops a single trade or a brief spike, not a book held away for minutes.

Worked example. The book sits 12% below an index of 36.60 and stays there. The mark follows M2:

A long opened at 36.60 with 5× leverage has its liquidation price near 32.53, 11.1% lower. This path crosses it after about six and a half minutes, with the index unchanged.

Funding on BVIV

Funding on most perps pulls the contract toward spot. BVIV has no spot. Funding here charges the carry built into a rolling 30-day volatility index, and adds the market's own premium over the index.

Funding has two parts, both standard HIP-3 mechanics:

1. An interest component, the modelled carry of the index. The oracle computes it every minute from the Volmex volatility curve, and Kinetiq relays it on-chain with setFundingInterestRates.
2. A premium component, produced by this market's order book relative to the oracle. It is sampled every five seconds and averaged over the hour.

A third input, the funding clamp, sets how far the interest component can pull the rate away from the premium.

Refer to the hyperliquid documentation for the latest funding calculations (https://hyperliquid.gitbook.io/hyperliquid-docs/trading/funding⁠)

Example. The interest relayed on 26 September 2026 was I = 0.2333% per 8 hours, or 0.0292% an hour:

The funding clamp at launch and after

The clamp starts at 3 basis points per 8 hours (0.03%). It will be raised gradually to 30 basis points (0.30%). Step sizes and dates are not published in advance; each increase is announced beforehand on X and in these docs.

What each stage means for you:

• While the clamp is below the modelled carry. On 25 and 26 September 2026 the carry ran at 0.23–0.32% per 8 hours, far above 3 bp. With the book on the index, longs pay only the clamp, 0.03% per 8 hours, not the carry. The full carry is paid only if the market trades at a premium of about I − c over the index, 0.20–0.29% at those levels. Market makers who price in the modelled drift will quote that premium. The mark price then sits above the oracle by a similar amount, and a buyer pays the premium in the price.
• Once the clamp comfortably exceeds the carry. Funding equals the carry for any hour whose premium is within c of it, and the premium pulls less on funding.
• When the carry exceeds the clamp. Funding at a flat book is c, not I. A premium below I − c lowers funding one-for-one. A premium above I − c brings funding up to I, and no further until the premium passes I + c. In testing with a 30 bp clamp, the carry exceeded the clamp at 4 of the 20 hourly settlements between 16:00 on 25 September and 11:00 on 26 September, reaching 0.315% per 8 hours.
• Each change of clamp changes what you pay, even if nothing else moves. The clamp in force can be read on-chain: Hyperliquid's perpDexs info request lists it for mkts:BVIV under assetToFundingClamp, and an asset with no entry there uses the 3 bp default.

The interest component: modelled carry

Where the carry comes from. A 30-day volatility index rolls forward continuously. Each moment it picks up a slice of expected variance at the far edge of its window and drops one at the near edge. When the volatility curve slopes upward, the slice gained is larger than the slice lost, so the index is expected to drift up. When the curve slopes down, it is expected to drift down. That roll is valued from the 7, 14, 30 and 60-day Volmex prints. Quoting in volatility points rather than variance also carries a convexity cost, estimated from the index's own daily moves. The carry is the roll minus that cost.

Without funding, a long would collect that drift for free. The interest component charges it, so that under the model, holding the perp is a bet on changes in volatility beyond what the curve already prices in.

r is the carry, per year. term1 is the roll along the variance curve. term2 is the convexity cost.

• B7, B14, B30, B60 — Volmex 7, 14, 30 and 60-day prints, as decimals (36.53 → 0.3653)

• v is the variance at the short end of the curve
• FV is the forward variance 30 days out

I is per 8 hours: 1,095 eight-hour periods a year.

Two points for traders:

• The sign is an output of the calculation. Funding can run either way, longs paying shorts or the reverse, depending on the shape of the curve and on the premium. It is not a view on direction.
• Funding moves when the curve moves, and the short end matters most. At the curve of 25 September 2026, one volatility point on the 7-day print moved I by about 5.4 bp per 8 hours in the opposite direction. One point on the 60-day print moved it by about 2.3 bp in the same direction.

Examples. Each row assumes the hour's premium sits inside the corridor, so funding equals I.

In the oracle’s worked example, a 10m USDC long pays about 188 USDC an hour. A flat curve still gives a negative carry, because the convexity cost remains when the roll is zero.

What the model does not do. The oracle’s implementation notes list these points:

• It does not re-sign early on a volatility jump or when the curve inverts. It flags a 14-day print above the 60-day print for monitoring only.
• It does not cross-check the prints against Deribit.
• It reports, but does not enforce, a step limit of 5 bp per 8 hours between publications. Its only hard limit is Hyperliquid's range of ±1% per 8 hours.
• It refreshes the convexity term once a day, at 00:00 UTC, from daily closes. If the history is unavailable at midnight, a live 30-day print within 180 seconds of midnight stands in and is marked provisional until the official close arrives. If the history cannot refresh at all, the stored variance is kept.
• It treats every year as 365 days. The 8-hour rate I is unaffected, but in a leap year a position held all year pays 366/365 of the stated annual rate, about 0.27% more.

The model's main assumption. The roll treats the volatility curve as an unbiased forecast of future variance. Option-implied variance usually includes a risk premium, so the curve's slope can overstate how much the index actually rises. Holders pay the modelled drift whether or not the index moves as modelled.

The hourly running average

Both parts of funding are averages over the clock hour, and both reset at the top of the hour.

The interest component is a running mean over the current UTC hour:

Each minute k of the hour:

• term2 is fixed for the UTC day
• I_k is published by the oracle, then relayed on-chain

At hh:00 the window closes, and the next minute starts a new mean.

• Relayed every minute, the value on-chain at the top of the hour is the average of that hour's samples, normally 60. That is the *I* used for the hour's funding. It is not a rolling 60-minute window. Nothing carries over from one hour to the next.
• Early in the hour the published I rests on a few samples and moves more. It settles as the hour fills.
• A missed minute leaves that hour with fewer samples. If updates stop across the top of the hour, Hyperliquid uses the last value that landed. In testing, the interest held for 8.7 minutes across a turn of the hour and the held value was paid.
• The oracle also publishes the previous, completed hour. A relay that updated only once an hour would have to use that value, which would apply each hour's carry one hour late.

The premium component is averaged the same way. Hyperliquid samples it every five seconds and averages the current clock hour's samples.

The predicted rate. During the hour, the funding rate shown on screen is built from the samples so far and the interest on-chain now. It becomes the payment only at the top of the hour, and early in the hour a few minutes of premium dominate it. In testing, a stress test held the book 10% above the index and brought it back at 18:38 UTC. The predicted rate stayed above 0.70% an hour until the 19:00 payment of 0.702%, then showed −0.048% almost four minutes into the next hour.

Example payment. A long of 1,000 contracts with the oracle at 36.50 has a notional of 36,500 USDC. If funding equals the carry of 26 September 2026, 0.0292% an hour, it pays 10.64 USDC an hour or 255 USDC a day. At 5× that is 0.15% of the position's margin each hour and 3.5% each day, deducted from its isolated margin. At the launch clamp with the book on the index, it would pay 1.37 USDC an hour.

Controls on the interest component

• Protocol. Hyperliquid accepts interest between −1% and +1% per 8 hours, clamps between 0 and 1% per 8 hours, and multipliers between 0 and 10.
• Oracle. The funding feed publishes nothing unless all four prints are present, fresh (under 180 seconds old) and in range, its stored series of daily changes is complete, and I is within ±1% per 8 hours. A refused execution leaves the last published value standing.
• Kinetiq. Pre-relay, the relayer applies only HL's ±1%/8h clamp.
• Multiplier. Fixed at 1. A multiplier scales both the interest and the premium. Zero funding does not halt trading or liquidations.
• Clamp. Set by Kinetiq and raised gradually, with notice, as described above.
• Notice. Changes to parameters are announced on X and in these docs, in advance where practicable. Changes in the modelled carry itself, including changes of sign, follow the volatility curve and are not announced.

Parameters and how they were chosen

Degraded mode and the internal oracle

Other mkts assets fall back to Kinetiq's internal pricing when their data source is closed, as described on Oracle prices. BVIV has its own fallback, built into the oracle’s price feed. While the Volmex index is fresh the fallback plays no part. When the index is stale, it can make this market's own order book the reference for the oracle, within bounds. This section discloses how.

Feed states.

Ages are judged in whole seconds, so a 180-second limit acts at 181 seconds. With the oracle’s default settings the staleness threshold and the hold budget are both 180 seconds.

The discovery rule.

On each update in discovery: target is this market's price on Hyperliquid, read from its impact prices (Hyperliquid's own mark while the book is unquoted).

6-hour time constant.

• step is at most 0.5% of the oracle
• oracle is kept within 5% of the anchor (anchor = last oracle set from a fresh print)
• at 90% of the band, re-anchor: the anchor moves to the band edge (once each way)
• with no re-anchor left, stop at 90–100% of the band

How far and how fast. Discovery is slow by design. It closes 2.7% of the gap to the book in 10 minutes, 15% in an hour, 39% in three hours and 63% in six. With the book 20% away it reaches the 5% band in about 1.7 hours, and with the book 10% away in about 4.2 hours. After its one re-anchor, the oracle can end up about 10% from the last valid print in one direction, at most 10.25% above or 9.75% below, and there it stops.

Recovery. When a fresh print arrives, the oracle returns to the index in one update. The jump is the index's move during the outage, less any discovery drift. The mark follows at no more than 1% per update, and the median keeps it steady while the book catches up. In testing, a +12.5 bp jump after a three-minute gap left the mark unchanged.

What this means for you.

1. During discovery, this market prices itself. The oracle, the premium measured against it, the notional that funding is paid on, and all three mark inputs come from mkts:BVIV's own order book, within bounds. The only venue-independent value left is the external price, which keeps showing the last Volmex print.
2. It can be moved. A participant who controls the book during a long outage can drift the oracle up to about 5%, or about 10% after the re-anchor, and the mark with it. That can trigger liquidations that would not happen at the index.
3. The return can be abrupt. Positions opened at discovered prices are revalued against the index as soon as it returns.
4. Funding keeps its last interest. The funding feed refuses stale prints, so no new interest is published during an index outage and the last relayed value stands. The premium is measured against a discovered oracle that converges on the book, so it tends toward zero, and funding tends toward the held interest, limited to the clamp: 3 bp per 8 hours at launch.
5. A relay outage is different. If no updates reach Hyperliquid at all, for any reason, the oracle freezes at its last value, there is no discovery, and after about 10 seconds the mark follows the local book alone.
6. If Hyperliquid cannot be reached when discovery is due, the oracle publishes nothing. The effect is the same as a relay outage.
7. A restart does not reset the state. The oracle keeps the feed's state between updates. A brand-new deployment during a long index outage would publish the last, stale print, and its mark basis would start at the full gap between the book and the oracle instead of building up over 150 seconds.
8. Kinetiq retains the right to halt the market, including during a long outage. Halting cancels all orders and settles every position at the current mark, which during an outage may itself come from the order book.

During discovery, the oracle’s defaults are shown. Setting the band to zero makes the oracle hold the last anchor instead of discovering. The anchor is whatever the oracle was when discovery started, which can be the last oracle price.

Fees

BVIV is not listed in growth mode. Growth mode is a HIP-3 setting that cuts trading fees on an asset by about 90%, and several assets on mkts use it. BVIV does not, so its fees are about ten times the growth-mode rates shown on Fees & costs.

With the deployer fee scale of 1.0 set on-chain, a HIP-3 trade pays Hyperliquid's base perp rate twice: once as the protocol fee and once as the deployer fee. At Hyperliquid's entry tier that is 0.090% for takers and 0.030% for makers, before staking and referral discounts. Trades through the Markets front end also pay the Markets builder fee.

Risks and disclosures

This section sets out some of the risks specific to this market. It adds to, and does not replace, the general risks of leveraged perpetuals on Hyperliquid and the disclosures for Third party assets.

The product and the index

• Leverage. At 5×, an 11.1% fall in the mark liquidates a long and a 9.1% rise liquidates a short. Over the 60 daily changes to 24 September 2026, the index's typical daily move (root mean square) was about 3.5%, so a liquidating move is about three typical days' moves in one. Volatility indices have fat tails, and moves of that size happen.
• Volatility spikes. Implied volatility can jump within minutes when bitcoin moves sharply, and its largest moves are usually upward. Shorts carry this gap risk: a spike can pass a short's liquidation price before the position can be closed, and the liquidation can fill well beyond it.
• No convergence. There is no spot to deliver or arbitrage. Nothing forces the perp back to the index except funding and the willingness of market makers to quote around it.
• A smoothed, model-built index. BVIV lags a sudden repricing of options by about its one-minute half-life, or two minutes around the 08:00 UTC expiry. When its replication cannot be computed, Volmex substitutes an estimate from at-the-money volatility and a tail measure.
• Two source venues. BVIV is built from Deribit and OKX options, with Deribit as the reference underlying price. An outage, error or manipulation at either venue passes into the index.
• The index belongs to Volmex. Volmex can revise the methodology, correct or restate values, or stop publishing. Kinetiq does not control or verify the index.

Funding

• Funding follows a model, not only the market. The carry is computed from the Volmex curve. On a flat or inverted curve it turns negative and shorts pay.
• It can change quickly and without notice. One volatility point on the 7-day print moves the carry by about 5 bp per 8 hours. The rate shown during an hour is an estimate. You pay the rate that stands at the top of the hour.
• The clamp decides how much of the carry you pay directly. At 3 bp, most of the carry is paid only if the book trades at a premium. As the clamp is raised, more is paid through the interest component.
• How funding responds to the premium depends on the clamp. While the clamp is below the carry, as at launch, funding is the premium plus c until the premium reaches I − c, so it follows the book one-for-one in both directions..
• Samples without a premium are dropped. When a side of the book holds less than 6,000 USDC, that five-second sample is left out of the hour's average.
• Funding comes out of your margin. Payments are taken from isolated margin every hour and move your liquidation price closer. If funding pays the full carry of late September 2026, a 5× long loses about 3.5% of its margin a day. At Hyperliquid's cap of 4% an hour ( to be updated according to Hyperliquid’s new terms ), a 5× position would lose 20% of its margin an hour. In testing, an hour in which the book spent about half its time 10% above the index paid 0.70% of notional.
• The model's assumptions may not hold. The carry treats the curve as an unbiased forecast. Implied variance usually carries a risk premium, so the modelled drift can exceed the index's actual drift. The convexity term uses 60 days of history and refreshes once a day.
• Limits can bind. If the market specific limits or Hyperliquid's ±1% per 8 hours bind, the rate on-chain differs from the model. A multiplier below 1 scales all funding, and a multiplier of zero switches it off.
• Update gaps. If interest updates stop, the last value stands, across hourly payments if necessary.

Mark price, liquidation and auto-deleveraging

Mechanisms inherit Hyperliquid infrastructure, please refer to
https://hyperliquid.gitbook.io/hyperliquid-docs/trading/liquidations⁠

Liquidity and order handling

• Liquidity may be thin and concentrated. Liquidity for third-party assets is arranged by the third-party deployer.
• Order price limits. At 5×, fills more than about 20% through the oracle are refused. In a fast move, the part of an exit that would fill beyond that is dropped.
• The open-interest cap. At the cap, new positions are refused and orders more than 1% through the oracle are rejected, even orders that only reduce a position. In testing this left the book one-sided within a minute. Whether liquidations fill normally at the cap during a real crash could not be tested.

Oracle, relay and infrastructure

• A chain of providers. Prices pass from Deribit and OKX to Volmex, then through the oracle’s network and data proxy to Kinetiq's relayer, then to Hyperliquid. A failure at any link stops or degrades updates.
• The internal oracle. During an index outage the oracle can move toward this market's own order book. See Degraded mode and the internal oracle.
• Hyperliquid. The market runs on Hyperliquid and inherits its risks: outages, bugs, API limits and protocol changes.

Deployer and parameter changes

• Kinetiq can change the maximum leverage, margin tables, the open-interest cap, the funding clamp and multiplier, and fees. It retains the right to halt, settle or delist the market. Changes are announced on X and in these docs, in advance where practicable.
• BVIV is a third-party asset. Volmex proposed it and Perps.inc brought it to mkts. See Third party assets.

Reference documentation

The BVIV index (Volmex). BVIV is constructed and published by Volmex. Its methodology paper sets out how the index is built from option and futures data across venues, how the exponential smoothing works, and how the tenors are defined.

• Volmex Implied Volatility Indices, methodology paper⁠
• BVIV index page⁠

Hyperliquid.

• HIP-3: builder-deployed perpetuals⁠
• HIP-3 deployer actions⁠
• Funding⁠
• Robust price indices⁠
• Margining⁠
• Liquidations⁠
• Auto-deleveraging⁠
• Fees⁠

*Disclaimer: This documentation is non-binding and may be updated from time to time. No representation or warranty is made as to the accuracy, completeness, timeliness, performance or fitness for any particular purpose of this documentation, any Volmex index (including the BVIV and EVIV indices), the underlying methodologies or any data, calculation or output derived therefrom. Index and oracle methodologies may be modified or discontinued at any time at Volmex’s sole discretion. To the maximum extent permitted by applicable law, Kinetiq, Perps.Fun, Volmex and their auxiliaries assume no liability arising from reliance on or use of this documentation, any Volmex index, any index methodology or any related data or output. All intellectual property rights in this documentation, the Volmex indices, the index methodologies and any related designs, data and materials remain with Kinetiq, Perps.Fun, Volmex or their auxiliaries, as applicable. As used herein, “Volmex” refers to Volmex Global and any of its affiliates. Any access to or use of the services offered by Kinetiq, Perps.Fun, Volmex or their auxiliaries, including Volmex Indices, is subject to the applicable terms of service.*