Perpetual Futures

A perpetual futures contract is a derivative providing continuous, margined exposure to the price of an underlying asset. It carries no expiration or delivery; instead, the contract anchors to a reference price through a funding mechanism, periodic payments exchanged between long and short positions according to its premium or discount. At a premium, funding is debited hourly from the margin of long positions and credited pro rata to shorts; at a discount, the flow reverses.

Because that premium or discount (the contract's basis) can be captured through a delta-neutral position in the contract and spot, any deviation creates an arbitrage incentive that draws the two prices back together, making the perpetual an efficient vehicle for synthetic exposure without delivery or contract rolls.

• Oracle price. The reference value of the underlying asset that the contract seeks to replicate. For Hyperliquid's crypto perpetuals, each validator publishes a weighted median of the leading spot venues' prices, and the chain adopts the stake-weighted median across them.
• Mark price. A robust estimate of the contract's fair value, used to margin and liquidate positions and evaluate order triggers. Hyperliquid takes the median of three estimates: oracle-linked, local order-book, and external perpetual mids, limiting the influence of any single price source.
• Funding rate. The anchoring mechanism that incentivizes the traded price to remain close to the oracle: payments exchanged in aggregate between position holders, calculated from the contract's premium together with a fixed interest component.

The design makes perpetuals unusually flexible. A venue can list a market without custody, delivery, or wallet support for the underlying, provided it can supply a reliable oracle and risk controls, since funding delegates price alignment to arbitrage. Traders gain continuous delta-one exposure in a single contract, avoiding fragmentation across expiries and strikes while using collateral more efficiently. Accordingly, perpetuals account for approximately 93% of crypto derivatives trading (Ruan and Streltsov, 2025), with the ten largest centralized perpetual exchanges turning over $86.2 trillion in notional volume in 2025 and the ten largest decentralized venues adding another $6.7 trillion (CoinGecko, 2026).

Real-World-Asset Perpetual Futures

The perpetual's success has so far been confined to crypto underlyings, while traditional access to delta-one exposure remains uneven across asset classes.

For single-name equities, the US currently lacks a broadly available listed instrument: delta-one exposure is restricted to over-the-counter swaps, reserved by statute for "eligible contract participants," a status that, for individuals, requires more than $10 million in discretionary investments (CEA Section 1a(18)). Volume concentrates instead in short-dated options, with same-day expiries carrying 59% of SPX options volume in 2025, roughly half of it retail (Cboe). And while commodities carry standardized, accessible futures, even these do not trade around the clock: outside listed hours, no instrument in the stack is available at all.

Crucial periods of benchmark price discovery therefore go untraded. During the Strait of Hormuz closures, crude reopened 11.5% and 17.0% higher on consecutive weekends, and this year the mean absolute weekend gap across major commodities is 1.8% against a 1.1% median, with WTI's 90th-percentile gap reaching 6%. The same concentration holds in equities, where 90% of earnings releases land outside regular hours and most post-announcement discovery completes within minutes (Pan, Sul, and Wang, 2026; Christensen, Timmermann, and Veliyev, 2025). Institutions can source weekend cover through bespoke swaps, negotiated dealer by dealer behind the same statutory gate; everyone else is locked out of the hours where price discovery concentrates.

Extending perpetual futures to traditional assets addresses both constraints. First, and most broadly, it opens standardized delta-one exposure to any asset and any collateralized participant, including exposures currently reserved for ultra-high-net-worth individuals. Second, because the contract trades continuously, participants can hedge and manage risk through the closures where price discovery concentrates.

The Architectural Design of RWA Perpetuals

Delivering on both, in practice, divides into two roles: an exchange layer, responsible for matching, margining, and settlement, and a contract specification layer, responsible for parameterizing the instrument itself, from its oracle and mark to its leverage and price bounds.

Hyperliquid: Open Exchange Infrastructure

The first requirement is an exchange layer capable of 24/7 operation. While traditional derivatives exchanges (i.e., “designated contract markets” or “DCMs”)  and clearinghouses (i.e., “derivatives clearing organizations” or “DCOs”) could in theory support around-the-clock trading, clearing, and settlement, they are bottlenecked by post-trade infrastructure: variation margin moves through settlement banks on the banking calendar, weekend collateral calls would raise liquidity demands weekday banking processes are not built to meet, and continuous operation requires surveillance and system resilience without maintenance windows (CFTC Letter 26-16, 2026). 

Hyperliquid solves these constraints by unifying execution and post-trade within a single protocol. HyperCore, the chain's state machine, provides matching, margining, and collateral control infrastructure as protocol logic, ordered by HyperBFT, a consensus algorithm run by a stake-weighted validator set. Three properties make the architecture suitable for 24/7 perpetual trading:

• Instant settlement. Execution and settlement are one event: every order, trade, and liquidation finalizes at consensus against prefunded USDC collateral, at 200ms of median end-to-end latency.
• Continuous risk management. Hyperliquid’s margin logic marks positions to settlement prices; HyperCore marks against a validator oracle refreshed every few seconds, accrues funding hourly, and liquidates through a deterministic waterfall that includes automatic order book liquidations, backstop liquidators, and, as a last resort, auto-deleveraging. It has margined continuously since early 2023, “clearing” over $4.7 trillion in volume through June 2026.
• Disintermediated access. Traditional derivatives access runs through layers of intermediaries, under terms negotiated per relationship. Hyperliquid users can access HyperCore on a disintermediated, non-custodial basis, on identical margin and terms.

Market deployment inherits Hyperliquid’s trade execution, clearing, and settlement infrastructure, sharing the same set of programmatically enforced rules. Under HIP-3, listing is delegated to third-party deployers, incentivized to provide secure and technically sound markets via a staking-and-slashing mechanism. HIP-3 deployers list markets against a bond of 500K HYPE, slashable by stake-weighted validator vote for faulty operation of their markets. The deployer then bids for listing capacity in recurring Dutch auctions, sets contract specifications and leverage limits, publishes oracles, controls listings and delistings, and may retain up to half of the fees its markets generate.

TradeXYZ: The Deployer Layer

TradeXYZ carries this mechanism to real-world assets. Launched in October 2025, it deploys HIP-3 perpetuals referencing offchain assets, with no tokenization or custody of the underlying. As of August 2026, the universe spans 103 deployed markets, 88 actively traded, across commodities, foreign exchange, US and Asian equities and indices, and pre-IPO products, including Cerebras (CBRS) and SpaceX (SPCX), both listed before their Nasdaq debuts, and ChangXin Memory (CXMT). Thirty-day daily volume averages $3.7 billion, cumulative volume exceeds $440 billion, and open interest stands at $3.5 billion; since July 17, TradeXYZ's seven-day volume has run above Hyperliquid's native crypto perpetuals, making the deployer's markets the largest on the exchange within ten months of launch.

As deployer, TradeXYZ defines the markets it lists, each deployment carrying an auction cost, and establishes risk parameters across that universe. Its most consequential function is adapting the crypto perpetual itself, an adaptation that reworks the three components defined above: i) the oracle, which carries the underlying's price, must remain continuous for assets that do not trade 24/7; ii) the mark, which carries the contract's fair value, must stay robust through temporary dislocations, so liquidations execute only against representative prices; and iii) funding must sustain the arbitrage incentive that anchors the contract while pricing each asset's cost of carry, keeping positions economical to hold.

TradeXYZ solves all three by splitting the contract's pricing mechanics into two sessions: external, while the underlying trades, and internal, while it is closed.

RWA Perpetuals: External Sessions

During the external session, when the underlying's venues are open, a TradeXYZ market functions as a crypto perpetual does.

In place of a crypto index, TradeXYZ's relayers publish the underlying's reference price feed, derived from its most liquid venues and chained across sessions for continuity: a US equity reference tracks Nasdaq through the regular day, extended hours after the close, and Blue Ocean (BOATS) overnight; index references combine cash indices with futures stripped of financing and expected dividends; commodity references smooth rolls between expiring contracts; and Korean equities convert through USD/KRW. To align carry with traditional assets, TradeXYZ halves the crypto perpetual's fixed interest component to roughly 5.5% annualized.

Market makers carry one added burden, hedging offchain through a broker, which duplicates collateral and introduces basis and transfer-latency risk. The process is otherwise identical: the contract, tethered to its underlying through funding, should create sufficient incentive to sustain performant, liquid markets.

To test how these markets perform empirically, we benchmark them against Hyperliquid's most liquid native-crypto perpetuals while external markets are open. The RWA perpetuals anchored tightly to the references they synthetically replicate, with a median mark-oracle basis of 4.31 bps against 5.08 for native crypto; 57.8% of minutes fell within 5 bps of the oracle against 49.1%; and even the tails staying close, 12.59 against 13.25 bps at the median market's p99 deviation from the underlying.

Similarly, despite the added overhead and nascent listings, quoted spreads ran tighter on the RWA markets, 1.67 against 1.88 bps at the median, and, like the basis, the tails stayed contained, 10.71 against 13.11 bps at p99. The tightness was matched in size, with the median RWA market-day displaying $211,000 of two-sided depth within 10 bps against $33,000 for the median crypto market, though BTC and ETH remain the venue's deepest books. As a result, RWA trades execute as liquid crypto does: the signed median fill lands 0.95 bps from the prior midpoint against 1.38, with 21.6% of orders price-improving against 15.8%.

Volume has followed that quality, and TradeXYZ's HIP-3 markets now carry a dominant share of Hyperliquid's perpetual volume, including eighteen of its twenty-five largest markets.

RWA Perpetuals: Internal Sessions

During internal sessions, when no external venue is live, TradeXYZ transitions the contract to an internal reference: the oracle and mark start from Friday's closing print and thereafter track Hyperliquid's own order flow. In open hours, funding computed against a live external price ties the perpetual to its underlying; through the closure, no such price exists, and the anchor becomes a scheduled event as the oracle reverts to external pricing at the reopen.

The contract therefore spends the weekend like a dated future, "settling" at the reopen to the recovered oracle print, as a future does at expiry. And because the market trades continuously, traders price that settlement in advance: WTI closes Friday at $100; the perpetual trades at $104 on Saturday evening; a trader expecting a $102 open shorts the contract and realizes the difference when the oracle resettles.

To balance the mark's flexibility against protection for positions carried across the closure, TradeXYZ introduced discovery bounds: a price range that limits how far an instrument can move within the internal period. The band is generally configured to 1/max-leverage, ±5% for WTI at 20x and ±2% for SP500 at 50x, ensuring that a position opened at the last external price cannot be driven past liquidation during weekend price discovery, even at maximum leverage. The band has since been adjusted for several markets, such as WTI, that reached their limits: under consistent pressure it may now re-anchor, up to twice for WTI, before hardening until the reopen, but the invariant remains a maximum deviation the mark cannot exceed.

And just as the RWA perpetuals proved robust in their external sessions, the internal sessions delivered similar results. Compared with each market's own main-hours baseline, weekend spreads ran at 85% of that level, no wider in 65 of 100 market-weeks, and the pooled distributions tighter at every reported quantile, 1.43 against 1.77 bps at the median.

However, while spreads held, depth thinned, retaining 50%, 41%, and 31% of its external level within 5, 10, and 25 bps. The shallower book came with smaller flow, the median order falling from $471 to $106, roughly four times smaller, yet typical fills still matched external sessions against the prior mark, 1.58 bps against 1.89, with the upper tail widening to 41.2 against 23.7 bps at p99 as larger orders encountered the thinner book, the natural cost of being the only venue open.
 

Still, for smaller trades and for traders willing to accept that wider tail, the sessions offer the first permissionless, order-driven weekend market in traditional assets.

Opening Auctions and the Value of Continuous Price Discovery

Having established the theoretical and empirical foundations of 24/7 perpetuals, we turn to what weekend trading does to the traditional markets it references. Weekend price discovery is strictly predictive: the internal price is a forecast of the opening print the market will settle toward, and once the underlying prints, the contract re-anchors to it, with 84% of reopenings back within 10 bps of the oracle inside five minutes and 89% inside fifteen. As a result, there is no channel through which the weekend price can directly disturb the open; the underlying's own prices govern the contract as soon as they exist.

For participants, the instrument replaces waiting for the reopen to absorb two days of information in one print, because the weekend tape prices it as it arrives: equities, with few weekend disclosures, realized 37% of their matched external-session volatility, while energy, moving on weather, outages, and geopolitics, held 64%. Any participant can put on delta-one exposure at a live price, and the convergence that protects the open completes the hedge, since a position carried into the reopen “settles” towards the underlying reference. A portfolio manager can cut equity risk on a Saturday, an options desk can manage delta while its listed hedges are closed, and a refiner can meet a supply shock at once instead of waiting to be marked.

For the wider market, the same price closes an information gap. The weekend is closed from Friday's close until CME's Sunday pre-open, whose indicative price arrives only in the hour before trading resumes, on exchange data feeds, and cannot be traded; Hyperliquid publishes the reference continuously, as a live, executable price anyone can read. This price feed, kept accurate by economic incentives, tells every market participant where the market is expected to open and gives them a basis for decisions, from risk marks to margins, ahead of the print.

Hyperliquid’s weekend market lets any participant express a view or adjust risk through the closure, in either direction and on margin, gives auction participants a continuously formed estimate of fair value before they commit orders to the uncross, and broadcasts that estimate publicly, carrying the closure's information beyond the venue's own participants.

Hyperliquid’s Weekend Prices against the Benchmark

With those functions established, Hyperliquid's weekend prices should be measurably better indicators of the open than the pre-closure boundary, as traders predicting the print pull the price toward it directly and hedgers positioning through the closure pull it indirectly. The prior sections showed the weekend market is orderly; improvement against the stale boundary would show it is informative, evidence that sophisticated participants trade these markets and that their prices carry signal.

To test this, we score two forecasts of the open against one target:

• Pre-closure boundary. The unchanged closing price. It uses no weekend information, so its error is the full move over the closure.
• Final weekend price. Hyperliquid's internal price, taken five minutes before the scheduled reopen.
• Opening reference. The target both forecasts are scored against: the median external-oracle print over the five to ten minutes after the reopen, past the oracle's handoff back to external pricing.

The panel takes one event per market per closure and requires $1 million of weekend volume, yielding 614 market-weekends across 68 markets and seventeen consecutive closures.

We find that Hyperliquid's weekend prices were generally accurate indicators of the open. The final weekend price finished closer to the reopening reference compared to the pre-closure boundary in 70.7% of the 614 market-weekends, and the share rose with the size of the move, 86% for reopening moves above 100 bps, 64% between 25 and 100, and 31% below 25. As expected, where weekend news opened a material gap against the pre-closure boundary the market priced it, while quiet weekends left little error to beat and no clear direction to call.

In addition to being directionally accurate, the weekend prices landed close in magnitude, measured by absolute error, the distance between each price and the reopening print. Median absolute error fell from 113.3 bps at the pre-closure boundary to 53.4 at the final weekend price, a 53% reduction; pooled mean absolute error fell 47%, from 169.2 to 90.4 bps; and the median event improved by 37.7 bps. 

The margin also grew with the move, the pooled median edge running +0.38 percentage points across all events and +1.22 among the 334 reopening moves above 100 bps. And it held weekend by weekend, positive in sixteen of seventeen closures across all events, and in all seventeen among the moves above 100 bps.

The reduction also held within every asset class, pooled MAE falling 12% in international equities, 55% in FX, 53% in US equities, and 41% in commodities and indices, although international-equity median error rose from 229.4 to 287.7 bps.

As trading was continuous and accumulated over the weekend, median error declined at every checkpoint, 113.3 bps at the boundary, 89.9 a day out, 70.1 six hours out, and 53.4 at the final five minutes, while the share of events closer than the boundary rose from 56.7% to 70.7%, and from 66.2% to 85.9% among moves above 100 bps.

Hyperliquid's pre-reopening prices were, in sum, informative. Across the seventeen closures, they removed 484 of the 1,039 percentage points of reopening error the boundary accumulated, with 54 of 68 markets ending the sample ahead and the deepest error pools cut the hardest.

Regressing the opening move on the implied weekend move, to understand whether priced moves arrive at full size, we find that 100 bps priced over the weekend carried into about 83 bps at the reopen, with the implied move alone explaining 69% of the variation. The largest gaps were priced most completely, 49% of a 100–200 bp gap left unpriced against 19% above 500 bps, and the relationship held market by market, positive in all 31 established markets at a median slope of 0.87.

Finally, to make sure these results are not influenced by external factors, we ask which way information flowed in the hour before trading resumes, when CME's Sunday pre-open accepts orders and calculates an indicative opening price from pending interest. If market makers quoted Hyperliquid off that print, the final hour's accuracy would be inherited and a repricing would mark the indicative's arrival. Across the 74 futures-linked reopenings, median error was 57 bps at T-65, the last checkpoint before order entry begins, against 59 at T-5, and the median event moved 0.3 bps through the hour, with no repricing when the indicative posted. 

Hyperliquid's prices had, over the weekend, incorporated enough information that they held their level even once the exchange's indicative opening price appeared.

Conclusion

In our December 2025 report on equity perpetuals, we argued that equity perpetuals, and RWA perpetuals more broadly, would become a defining theme of 2026. Eight months later, TradeXYZ's trailing 30-day volume has risen from $6.9 billion to $110.6 billion, a sixteen-fold increase, and it now accounts for 52% of all Hyperliquid perpetual volume, up from 4.3%.

This growth, although substantial, was not unexpected; it was telegraphed by Hyperliquid's zero-to-one improvements in market infrastructure. Hyperliquid simplifies market creation by providing exchange, clearing, and settlement infrastructure that new markets inherit rather than rebuild. Market rules live in onchain logic and the protocol enforces them, so every deployment starts with the same institutional-grade foundation. Non-custodial frontends enable a disintermediated market structure: users control their own wallets, and no intermediary ever stands between a trader and the market.

Deployment separates along the same lines, splitting functions traditionally bundled inside a single exchange between HIP-3 deployers, who design and launch markets, and the Hyperliquid layer itself, which provides neutral execution, clearing, and settlement that no single party controls. Trading and settlement systems have traditionally been the province of specialized technology vendors, which build and license them to exchanges and clearinghouses at a premium. Hyperliquid makes that same infrastructure a public utility: deployers, interfaces, and even registered exchanges can plug into the protocol and inherit its matching, clearing, and settlement.

The results of the weekend markets and RWA perpetuals are therefore a reflection of broader dynamics, of what happens when global participants can provide, price, and manage risk on one layer with one set of rules rather than across fragmented systems. We believe the trend continues, from the pre-IPO listings of today to future HIP-4 markets, as permissionless, performant systems built for broad access keep extending, asset by asset, into the hours, markets, and products traditional structure has left unserved.

Appendix

A.1 Data Collection

The analysis draws on 0xArchive Hyperliquid data covering June 13 through July 13, 2026: the twenty-five largest TradeXYZ perpetuals and the twenty-five largest native-crypto perpetuals by trailing volume. Six series are used for the analysis:

Session status follows each contract's specification: the oracle tracks the external reference while the underlying trades and advances on internal pricing while it is closed. Because those hours differ by asset, cohort comparisons use fixed windows inside the official sessions. The external window runs Monday through Thursday, 14:00–16:00 UTC, June 15 through July 9, covering the twenty-three non-Korean TradeXYZ markets and all twenty-five crypto markets; SKHX and SMSN are measured separately on their Korean session, Tuesday through Thursday 00:00–06:00 UTC. The weekend window runs 12:00–20:00 UTC on the four Saturdays, June 20 through July 11, hours that are internal for every TradeXYZ market, with the crypto cohort as a clock-matched control. Every comparison below runs inside these windows.

A.2 Mark-to-Oracle Basis

For the mark-oracle comparison, we define the signed basis as 10,000 × (Mark − Oracle) / Oracle: positive where the mark stands above its oracle, negative where it stands below, with rankings and tail statistics taken on the absolute value. The basis measures the integrity of the contract's price stack, since the oracle carries the underlying's reference price and the mark is the value against which positions are margined and liquidated.

The common panel pairs the twenty-three non-Korean TradeXYZ markets with the twenty-five crypto markets across sixteen shared U.S.-clock dates, 768 market-days; adding SKHX and SMSN on their own Korean clock contributes twenty-four more, giving the broad fifty-market view 792 market-days and 100,800 one-minute observations.

The anchoring is uniform across markets. Median absolute basis sits below 9 bps in 48 of 50, p95 stays under 20 bps in 45, and the cohorts are interspersed through the ranking, TradeXYZ holding eight of the ten tightest. The exceptions are the Korean conversion listings, SMSN and SKHX, whose oracles convert KRX prices through USD/KRW and whose medians of 26.8 and 30.7 bps place them last; for CBRS and PUMP the box reaches zero, at least a quarter of their minutes printing the mark exactly at the oracle.

On the common clock, TradeXYZ anchors slightly tighter at the center, the equal-market-day median at 4.31 bps against 5.08 for crypto and 57.8% of its minutes within 5 bps of the oracle against 49.1%. The signed distributions separate the cohorts more clearly than the absolute ones, TradeXYZ marking modestly above its oracle where crypto sits modestly below, and the tails are comparable, 12.59 against 13.25 at the median market's p99.

The broad view separates only where Korea enters. With SKHX and SMSN included, TradeXYZ's p95 and p99 rise to 19.0 and 52.2 bps against 12.9 and 20.7 for crypto; restricted to the common clock, TradeXYZ's p99 falls back to 18.46 bps, below crypto's 20.67, so the heavier tail belongs to those two listings rather than the cohort.

When deviations occur, they are brief and isolated. Nearly nine in ten market-days touch above 5 bps at least once, yet only 5.5% contain any run above 25 bps, 20.7% hold above 10 bps for five consecutive minutes, and the median session sees a single market above 25 bps. The one clustered episode belongs to crypto, nineteen of its twenty-five markets running above 25 bps in the same June 25 session, with no comparable shared date on the TradeXYZ side. 

Overall we find that the two cohorts anchor almost identically on the common clock, TradeXYZ slightly tighter at the center with an equal-sized premium to crypto's discount; the full statistics, including the Korea-widened broad view, are below.

A.3 Quoted Spreads and Displayed Depth

For quoted liquidity, we define the spread as (best ask − best bid) / midpoint × 10,000 on the one-minute BBO grid, and displayed depth as the conservative two-sided notional: the smaller of the bid and ask sides within 5, 10, and 25 bps of the midpoint, so a book counts as deep only where it is deep in both directions. Depth comes from full-depth checkpoints built from each market's complete L4 resting-order stream, sampled on the strictly prior five-minute clock, with hourly checkpoints retaining the five-largest-markets-per-cohort ladder study. The common-clock panel contains 768 market-days across 23 non-Korean TradeXYZ markets and 25 native-crypto markets.

Median quoted spreads sit below 2 bps in 22 of 50 markets and below 5 in 46, and the tightest books mix the cohorts: SP500 quotes 0.13 bps at the median, inside SOL at 0.14, BTC and HYPE at 0.16, and SILVER at 0.17. The four widest medians are CBRS, JTO, PUMP, and NBIS, two of them crypto. Unlike the basis ranking, the Korean listings sit mid-pack, SKHX at 2.57 bps and SMSN at 4.24: their wider basis does not carry into the quoted book.

Pooled, TradeXYZ quotes tighter from the median outward, and the gap widens through the tail. The equal-market-day median is 1.67 bps against 1.88, the p99 is 10.71 against 13.11, and 90.2% of TradeXYZ minutes quote within 5 bps against 83.1%. The one edge crypto keeps is at the touch itself, 37.0% of minutes inside 1 bp against 36.2%; everywhere beyond, TradeXYZ is tighter. Korea barely moves the broad view, the including-Korea median at 1.77 bps and the median market's pooled p99 holding at 5.87 against 7.06.

Panel-wide, the median TradeXYZ market-day displays $211,380 of two-sided depth within 10 bps against $33,055 for crypto; at the very top the order reverses, BTC at $43.0M and ETH at $23.7M within 25 bps on the raw checkpoint ladders against $17.8M for SP500 and $15.3M for XYZ100, with the top-five medians favoring crypto $3.61M to $2.21M, 1.6x, and SKHX again trailing its cohort at $454K. TradeXYZ's liquidity advantage is its breadth across listings; crypto's is the depth of its two largest books.

A.4 Execution

For execution, we score every non-liquidation reconstructed order against two references and define the cost as the side-adjusted distance, Side × 10,000 × (Fill / Reference − 1), positive where the taker filled through the reference and negative where it filled inside it, so price improvement reads as negative cost. The references are the strictly prior BBO midpoint, at most ten seconds old, and the strictly prior oracle print, in practice 3.09 seconds old at p99. Orders are reconstructed by grouping each aggressor's consecutive fills, and every statistic weights active market-days equally. The panels are large: 3.59 million midpoint-matched orders, 76.5% of TradeXYZ flow and 97.4% of crypto's, and 2.12 million TradeXYZ orders scored on the oracle.

Against the midpoint the cohorts fill almost identically, with TradeXYZ slightly tighter everywhere: the signed median is +0.95 bps against +1.38, 37.1% of fills land within a basis point against 34.6%, and 21.6% price-improve against 15.8%. The distributions separate only deep in the tail, 23.61 against 31.86 bps at the absolute p99.

Order size moves the tail more than the typical fill. The conditional median holds between roughly 1.4 and 3 bps across six orders of magnitude of notional, while the conditional p90 climbs from about 6 bps for the smallest orders into the mid-teens for six-figure crypto fills, TradeXYZ's tail flatten through the institutional range. Estimates render dashed where fewer than one hundred active days support them.

Because the contract margins against its mark and anchors to its oracle, we also score TradeXYZ fills on the oracle directly: the signed median is +1.79 bps, 79.6% of fills land within 10 bps, and 41.9% price-improve.

Execution statistics describe matched, realized fills, gross of fees: orders without a sufficiently fresh prior quote, (23.5% of TradeXYZ flow against 2.6% for crypto) are excluded, and submitted-but-unfilled size is unobservable. The crypto oracle column stays empty because the delivered series prints once a minute against every three seconds for TradeXYZ, too coarse to score individual fills.

A.5 The Weekend Internal Session

For the weekend, the comparison is TradeXYZ against itself: the four Saturday windows, June 20 through July 11, 12:00–20:00 UTC, scored against the same market's external sessions, with no crypto benchmark. Spreads pair each market-week with its own main-hours baseline across 100 market-weeks; full-book depth uses the same within-day-median, equal-weight design across 92 paired non-Korean market-weeks. Execution scores every non-liquidation order against the strictly prior mark, the reference that stays equally fresh through the closure, 3.1 seconds old at p99 in both sessions, across 3.23 million external orders and 401 thousand internal.

Quoted spreads do not widen through the closure: the median market-week quotes at 85% of its main-hours spread, 65 of the 100 pairs are no wider, and every summary quantile is tighter internally, 1.43 against 1.77 bps at the median and 7.90 against 10.98 at p99. The exceptions sit at both ends: SP500, the tightest book in the sample at 0.13 bps in main hours, widens to 2.5x across its weekends, while CL and BRENTOIL quote at roughly a third and half of their main-hours spreads.

However, depth is where the internal market thins: the median non-Korean market-week retains 50% of its own external-week conservative depth at 5 bps, 41% at 10 bps, and 31% at 25 bps; across the three bands, the middle half of pairs lies roughly between 24% and 64%. 

While the typical weekend fill is slightly tighter, 1.58 against 1.89 bps at the absolute median, the tail widens to 41.20 against 23.67 bps at p99, with the distributions crossing between the median and p90, 9.78 against 8.67 bps.

Pooled median order also falls from $471 to $106, and 84.6% of internal orders sit below $1,000 against 65.4% externally. Within bands, the smallest weekend orders fill slightly tighter than in main hours and costs diverge with size, to roughly 6 against 4 bps at the median above $100K: the tighter aggregate median owes to that mix, while the widening above $10K carries the tail.

Throughout, the weekend columns compare TradeXYZ with its own external baseline; execution scores against the contract's own mark while the underlying is closed.

A.6 Weekend Price Action

For the weekend session, we test whether prices formed while the underlying is closed predict the price that prevails when it reopens. The design makes this the traders' objective: the oracle resettles at the reopen, so a weekend position is in effect a forecast of the reopening price, and the weekend price aggregates those forecasts. The benchmark is the unchanged pre-closure boundary, which uses no weekend information, so its error equals the move over the closure. The target is the reopening reference, the median external-oracle print over the five to ten minutes after the reopen, past the oracle's handoff back to external pricing. Weekend prices carry information only to the extent they consistently beat the benchmark against that target.

The data run from March 13 through July 3, 2026, covering seventeen consecutive closures: thirteen ordinary cohorts and four containing market-specific schedule changes for Good Friday, Memorial Day, Juneteenth, or July 3. The first three months are stored on a five-minute grid, each market's internal price through the closure plus the external oracle on either side of it; the final month is the delivery's native three-second mark, midpoint, and oracle series. Close and reopen times follow TradeXYZ's instrument specifications, verified against the 2026 holiday schedule: Memorial Day is treated as an extended closure for U.S.-session equities: the boundary is Friday, May 22 at 20:00 ET; T-24h is Sunday, May 24 at 20:00 ET; T-6h is Monday, May 25 at 14:00 ET; T-5 is Monday at 19:55 ET; the reopen is Monday at 20:00 ET; and the reopening reference is Monday at 20:05 ET (Tuesday 00:05 UTC).

We construct one event per market per closure. An event enters the broad panel when the market remains listed, reports at least $1 million of weekend volume, and has complete observations at the boundary, all three checkpoints, and the reopening reference. This yields 614 market-weekends across 68 markets; 18 markets appear in all 17 closures, and each event carries equal weight. Two further panels serve specific tests: 74 reopenings in the four futures-linked contracts, CL, gold, silver, and SP500, across 20 closure dates for the pre-open analysis, and 100 reopenings across the four recent closures, with granular data, for re-anchoring and benchmark sensitivity.

A.7 Weekend Price Absolute Error

The first test is binary: whether each weekend price finished closer to the reopening price than its pre-closure boundary. It did in 70.7% of the 614 market-weekends. The share rises with the size of the reopening move, 86% above 100 bps, 64% between 25 and 100, and 31% below 25, where there is little error to beat, and every asset class clears parity, from 56% in international equities to 74% in US equities.

The margins behind those wins are large. Median absolute error falls from 113.3 bps at the pre-closure boundary to 53.4 at the final weekend checkpoint, a 53% reduction; pooled MAE falls 47%, from 169.2 to 90.4 bps; and the median event improves by 37.7 bps.

Both results hold within every class. Pooled MAE falls 12.4% in international equities, 55.4% in FX, 52.5% in US equities, and 41.3% in commodities and indices, although international-equity median error rises from 229.4 to 287.7 bps.

Disaggregated to the 68 individual markets, the result follows history length: all 31 markets with at least ten observed closures reduce pooled MAE, 23 of the 37 with shorter histories do, and 53 of 68 improve at the median error. The exceptions are markets where one or two weekends decide the record.

The remaining question on the broad panel is when those gains arrive. Repeating the comparison at each checkpoint shows median error declining monotonically, from 113.3 bps at the pre-closure boundary to 89.9 at T-24h, 70.1 at T-6h, and 53.4 at T-5, while the share of improved events rises from 56.7% to 70.7%. Accuracy accrues through the closure rather than at either end.

Direction is right in 88.0% of the 517 events where the reopening move reaches at least 25 bps, and sharpens with size: where the final weekend move reaches 100 bps, 50.7% of the sample, the sign is right in 94.9% of events. The improvement extends into the tail, with expected shortfall 38% lower across the worst decile of events, 36% across the worst 5%, and 39% across the worst 1%.

We re-estimate the reduction under stricter samples. Excluding the four holiday boundaries yields 45.1%; requiring at least $10 million of weekend volume yields 53.2%; the balanced 18-market panel yields 54.3%. A randomization test that reassigns targets within closure and asset class, preserving the common weekend shock, produces a null median of -1.4% against the observed 46.6%, with a one-sided p-value of 0.0001.

A.8 Linear Regression

Finishing closer says nothing about size: a market could beat the boundary every weekend while moving half as far as its underlying, or twice as far. The regression tests size directly, pairing each weekend's final internal move with its reopening move, both measured from the pre-closure boundary, one observation per market-weekend, uncertainty computed on the seventeen closures. If weekend prices are calibrated, the slope is one, each basis point priced over the weekend appearing at the reopening reference in full.

The full panel comes in just under that bar. The slope is 0.834, so a weekend move 100 bps larger comes with a reopening move about 83 bps larger, and the weekend move accounts for 69% of the variation in reopening moves. The intercept, −15 bps, is small next to the 169-bp average boundary error, and the interval consistent with the data, 0.73 to 0.93, sits below one.

The relationship strengthens with move size. The median T-5 miss falls from 49% of the reopening move for 100–200 bp events to 40% at 200–300 bps, 30% at 300–500 bps, and 19% above 500 bps. Larger dislocations are therefore not only anticipated in direction; more of the eventual move is already priced before the external reference returns.

The fit also strengthens through the weekend. The same regression at the earlier checkpoints gives slopes of 0.68 at T-24h, 0.82 at T-6h, and 0.83 at T-5, with explained variation rising from 37% to 54% to 69%.

Estimated market by market, the relationship is broad rather than carried by a few large names. All 31 markets with at least ten observed closures have positive slopes, with a median of 0.87, and 29 of the 31 are statistically distinguishable. Among the shorter histories, 16 of the 18 markets with at least three closures are positive as well; the two exceptions, BB and COST, rest on five and three weekends. The remaining 19 markets have fewer than three closures and are not estimated.

A.9 The Pre-Open Hour and Re-Anchoring

We then test whether accuracy improves during the final hour before the reopen, when CME accepts and nets orders and publishes indicative prices from pending interest; accuracy gained in that hour would be attributable to the venue's pre-open process rather than to weekend trading. On the 74 futures-linked reopenings we score T-65, the last checkpoint before order entry begins, against T-5 on the same post-reopen target. We find no incremental accuracy: the T-65 median error is 56.78 bps against 58.72 bps at T-5, the median paired gain, T-65 error minus T-5 error, is +0.30 bps, 37 events improve against 34 that worsen and 3 ties, and the closure-level Wilcoxon p-value is 0.872. Most accuracy is in place before order entry begins.

Finally, we measure convergence after each recent reopening at native cadence. Of the 100 reopenings, 84% re-anchor within 10 bps inside five minutes and 89% within fifteen. Excluding Korea, the final internal mark misses the reopening reference by 45.1 bps at the median, with a +13.1-bp median improvement over the frozen close and 60.9% of events improving.

The information contained in this report and by Blockworks Inc. and related affiliates is for general informational purposes only and is not intended to provide legal, financial, or investment advice. The report should not be construed as an offer or solicitation to buy or sell any security, token, or financial instrument and does not represent any recommendation or endorsement of any investment or financial product or service. Blockworks Inc. and related affiliates are not registered as a securities broker-dealer or an investment advisor in any jurisdiction or country.