Apparently, we, crypto people like to brag a little too much about the fact that Hyperliquid lets you trade a wide range of traditional assets during weekends.

So much so that I've been talking about it to a good friend of mine who works in commodity trade finance at UBS (he finances physical oil traders). He pinged me on Friday night asking me how predictive are prices on there over weekends. I realized a lot of people have been asking themselves the same question and as big commodities/equities traders on HL, we have the data (and also some help from @hydromancerxyz). So I decided to publish this article and finally answer the question everybody's been asking:

How informed are traders on Hyperliquid?

The data from this weekend

Let's detail the data from this weekend before running a wider analysis on previous weekends. We'll begin with perhaps the star of the show: oil. There are currently 4 main ways to trade oil on Hyperliquid during weekends: the CL contract deployed by @tradexyz, the OIL contract from @felixprotocol, the USOIL contract from @kinetiq_xyz and the ENERGY contract from @ventuals. They're all slightly different, CL and OIL are derived from WTI the US crude oil future, USOIL is derived from the US oil fund (USO), and ENERGY is the energy sector ETF XLE from State Street which is heavily correlated to WTI even tho it has other dependencies to Natgas and some equities (it's traded at a 0.6 beta to WTI which is high enough to include it in the analysis).

The CL and OIL contract got hit by the max limit move pretty fast, on the 28th at 7am UTC both contract were trading at the upper limit. It limited price discovery on Sunday, but we can still see how fast all those markets catched the first bombings. Here is the story, all timestamps UTC.

06:13:59: Everything is dead. It's 1:14 AM on the US East Coast. CME closed 8 hours ago. BTC on Binance: $65,580.62, trading 2-4 trades per second. CL on XYZ: $68.71, flat. USOIL on Kinetiq: $82.81, flat. OIL on Felix: $68.51, flat. ENERGY on Ventuals: $56.80, flat.

06:14:01: CL ticks up. Mid moves from $68.71 to $68.74 (+$0.03, +0.03%). Three cents. The ask lifts to $68.76. On Binance, BTC is dead 2 trades this second, price unchanged. Nobody else has moved.

06:14:21: CL jumps again. Mid leaps to $68.80 (+0.12%). The ask moves up to $68.88 the spread widens from 5 cents to 17 cents. Someone is lifting offers, probably noise given the subsequent trades. BTC on Binance: still $65,580.62. Still 2-3 trades per second. USOIL, OIL, ENERGY: all flat.

06:14:34: BTC drops on Binance. 607 trades in a single second. 5.4 BTC of volume. Price crashes $54 to $65,526 (−0.08%). This is the information arrival on the most liquid crypto market in the world.

06:14:35-36: BTC continues selling. −$65 (209 trades), then −$71 (136 trades). CL holds steady at $68.80.

06:14:50: BTC second wave. 776 trades, 6.6 BTC dumped in one second. Price drops to $65,467 (−0.17%). CL ticks up again to $68.82 (+0.15%).

06:15:00 to 06:15:35: CL sits at $68.80, flat. BTC keeps bleeding: −0.13% at 06:15:04, −0.19% at 06:15:18, −0.20% at 06:15:31. USOIL still hasn't moved stuck at $82.86. OIL and ENERGY not moving.

06:15:36: USOIL (Kinetiq) wakes up. First real tick: mid jumps to $82.92 (+0.13%). 62 seconds after BTC, 95 seconds after CL's first move.

06:15:38: CL accelerates. Ask lifts to $69.02, mid hits $68.87 (+0.23%). Someone is stacking orders on the offer side.

06:15:40: CL at $69.00 (+0.42%). Bid fills up to $68.99. In 6 seconds CL has gone from +0.13% to +0.42%.

06:15:44: USOIL jumps to $83.23 (+0.50%). A massive 26-cent ($0.26) move in 8 seconds. Simultaneously, BTC is getting obliterated 1,351 trades this second, 11.5 BTC dumped, price at $65,423 (−0.24%).

06:15:49: CL spikes to $69.16 mid (+0.66%). The ask side is $69.34, 48 bps spread.

06:16:02: BTC takes another leg down: 1,138 trades, −$214 (−0.33%).

06:16:07: CL touches +0.93%. Mid at $69.35.

06:16:25: CL breaks 1%. Mid at $69.44 (+1.05%). It's been 2 minutes and 24 seconds since BTC first dumped. CL is already pricing in a significant geopolitical move on a DEX, on a Saturday morning, with no CME for reference.

06:16:30: OIL (Felix) moves. First tick to $68.59 (+0.11%). But the book is thin: bid $68.03, ask $69.15 a 163 bps spread.

06:16:37: BTC collapse intensifies. 1,035 trades, 10.2 BTC, price down to $65,279 (−0.46%).

06:17:06: BTC at −0.55%. 1,541 trades this second. CL is at +0.84%, digesting near $69.29. USOIL is running higher.

06:18:21: BTC touches −1.5%. CL is near +0.84%. USOIL has run to ~+1%.

~06:24:30 — ENERGY (Ventuals) wakes up. 10 minutes and 16 seconds after BTC. The first prints are chaotic initially the mid drops to $56.36 (−0.78%), the wrong direction. Twenty seconds later at 06:24:50 it whips to $57.43 (+1.11%). Then back to −0.35% at 06:24:59. This is what price discovery looks like on a thin HIP-3 perp.

A few 10s of seconds were enough for OIL markets to move on a weekend. There is still some alpha to capture here if you're faster.

Now more names. Across 35 HIP-3 instruments spanning single-name equities, commodity futures, index ETFs, thematic baskets, FX, and bonds, Hyperliquid's weekend markets predicted the direction of the Monday/Sunday opening gap with 100% accuracy (34 out of 34 assets with a meaningful gap). The regression line through the data has a slope of 1.06 and an R² of 0.973, meaning HIP-3 prices explain 97.3% of the variance in the actual opening gap. The median prediction error at the moment of the reference market open is 14 basis points.

The precision varies sharply by platform and asset class. XYZ's single-stock perps are the standout: TSLA closed the weekend 3 basis points from the overnight open (-3.04% vs -3.01%), AMD landed within 3 bps (-3.16% vs -3.19%), and NVDA within 12 bps. Across 16 single names, the median error is just 13 bps, these are the tightest, most liquid HIP-3 books, and it shows. Kinetiq's index ETF perps are even more impressive on a per-asset basis: US500 (SPY proxy) was 4 bps off, USTECH (QQQ) 6 bps, SMALL2000 (IWM) 14 bps. Ventuals' thematic baskets also converge cleanly their SEMIS perp (SMH proxy) finished 1 basis point from the actual open, their ENERGY perp landed within 26 bps of XLE though DEFENSE overshot by 135 bps, likely reflecting a thin book overreacting to the war headline. For CME commodities, Gold is the crown jewel: XYZ's GOLD perp finished within 1 basis point of the CME gap (+2.23% vs +2.22%). Silver was 34 bps off, Platinum 95 bps, and Copper 107 bps the error scaling directly with how thin the HIP-3 book is for each metal.

The three visible outliers tell a structural story for USOIL and NATGAS, and a wide overprice for USOIL into the fear.

The hidden alpha is in the cross-deployer spread. TSLA trades on four platforms simultaneously (XYZ, Kinetiq, Felix, Cash), and all four landed within 4 bps of each other at Monday's open but during the weekend, they diverge widely due to funding rates. Whoever arbitrages these deployer-to-deployer lags on a Saturday morning buying the slow book, selling the fast one, captures a structural edge that exists purely because of fragmented liquidity across independent HIP-3 market makers. The chart's regression line doesn't show this: it shows the destination was right for everyone, but the path was wildly different, and the path is where the money is.

The final comment is about the opening. That's probably where you can scoop the most alpha since it's when the primary market meets back with Hyperliquid. And it's a pretty toxic relationship. For Gold and Silver at the CME Sunday open (23:00 UTC), the oracle gaps immediately to reflect the new CME print while the perp mid remains sticky at the pre-open level, creating a premium of +193 bps (Gold) and +292 bps (Silver) that decays to zero within 5–6 minutes. For PLTR at the overnight equity open (01:00 UTC), the effect reverses: the oracle updates first and the perp mid overshoots on the subsequent move, briefly trading 30–70 bps below the oracle during the initial rip before flipping to a +30–45 bps premium as the perp runs ahead. In both cases, the pattern is the same a temporary dislocation between perp and oracle that mean-reverts within minutes, suggesting a systematic short-term alpha in fading the opening premium across HIP-3 markets.

A more detailed quantitative analysis

I also ran a more broad quantitative analysis over more weekends. I analyzed 9 weekends across Gold (xyz:GOLD), Silver (xyz:SILVER), and NQ (xyz:XYZ100), comparing their weekend returns against the corresponding CME Globex opening gap on Sunday 23:00 UTC over 9 weekends in Jan-Feb 2026.

The results are quite impressive, with 90% hit rate on the direction of the weekend move. The 3 misses were small-gap weekends where the CME move was <0.05%.

The magnitude of the move is also decently informed. But a quite surprising fact about the magnitude of it is that the best indicator isn't at Sunday 23:00 UTC when CME opens, but around 20:00 UTC, when slope-to-CME sits near 1.0 (1.09 for Gold, 1.04 for NQ) and R² peaks at 0.73. In the final 3 hours, metals overshoot, Gold's slope jumps from 1.09 to 1.61, Silver from 1.88 to 2.04 while NQ stays pinned at ~1.0 throughout. The practical takeaway is twofold: HL weekend prices are genuinely informative for Monday's gap, but for metals, the signal peaks hours before CME reopens the last-hour move is noise, not edge.

The overshoot has a microstructural explanation. L2 snapshots show that liquidity providers pull 66–84% of top-5 book depth in the hours before CME reopens. Gold's book thins from $194k to $50k, NQ from $467k to $73k. Volume builds up a bit during those 3 hours, especially for GOLD with a 3x mean increase in the last hour before opening. The practical takeaway is that the cleanest HL-based estimate of the CME gap comes at 20:00 UTC, three hours before the market reopens, when the books are still intact and the price reflects pure weekend information not the microstructure impact of convergence trades into depleted order books.

Conclusion

So, how informed are traders on Hyperliquid? Very. 100% directional accuracy this weekend across 35 instruments, 90% hit rate over 9 weekends, and a median error of 14 bps at the open. The prices are real.

But the edge isn't just in knowing where Monday opens. It's in knowing when the signal is cleanest (20:00 UTC, not 23:00), where the liquidity cracks are (cross-deployer lags, book thinning), and how the perp-oracle dislocation trades at the open. The data is all there you just have to look at it on a Saturday.

Another interesting analysis could be made on fundings, as many design choices have been taken by the deployers, and it impacts market a lot.