One trade, many tickers:
correlation across Hyperliquid's markets

Daily-return correlation structure across 260 Hyperliquid markets — 176 crypto perps plus 84 tokenized equities, indices, commodities and FX on the xyz deployment — grouped into 39 thematic categories. Pearson correlation of daily log returns, 180-day primary window, 60-day recency check. Data through July 2, 2026.

Executive summary

How tight is each basket?

Average pairwise correlation between members of the same category, 180 days. High = the label is a real basket; near zero = the label is narrative, not risk.

Three reads. First, memecoins are a real basket — 0.67 average across 23 names (kBONK·kSHIB hits 0.91); they trade as a single flow. Second, hyperscalers are not — 0.29 average, with MSFT·ORCL (0.57, the AI-capex pair) the only strong link and GOOGL·IBM at 0.08. "Big tech" is not one leg. Third, several stock labels — commodities, healthcare, meme-retail, frontier-AI names — have essentially zero internal correlation: BRENTOIL·CL (0.90) is the only real commodity pair, while COPPER·NATGAS is −0.21.

Between categories: crypto

Equal-weight category indices, 180d. Blue = positive, orange = negative; deeper = stronger.

The blob is visible at a glance. The exceptions are what matter: the Hyperliquid ecosystem (HYPE + PURR) runs 0.4–0.57 to everything else — HL trades its own cycle; politifi (TRUMP, MELANIA, WLFI) sits near 0.6; and tokenized gold (PAXG) is 0.3–0.4 against all of crypto but 0.57 to the commodities basket — it is gold, not crypto.

Between categories: equities

AI semiconductors correlate 0.85 with the equity indices basket — semis effectively are the index now (SMH·SP500 is 0.78 at the symbol level). The memory-stock trade is a semis trade (0.60 to ai-semis), not a big-tech trade (0.20 to hyperscalers). The two neocloud names, CoreWeave and Nebius, form the tightest stock pair in the study outside indices and oil (0.79), each carrying ~0.5 NVDA beta.

Where crypto meets equities

Rows: crypto categories. Columns: equity categories. 180d, overlapping trading days only (~124 obs).

The pairs that matter

PAXG·GOLD at 0.76 is the internal sanity check — the same asset measured on a 7-day-a-week calendar against a 5-day one.

Focus baskets

Memory stocks

The US cluster (MU, WDC, SNDK) is coherent, and SK Hynix·Samsung is 0.84 — but Asia↔US pairs read low (KIOXIA·WDC 0.20) because Tokyo and Seoul close ~14 hours before New York. Same-date correlation structurally understates linkage for Asia names; treat those pairs as lead-lag relationships, not weak ones.

AI semiconductors

Hyperscalers

Crypto-linked equities

COIN·HOOD (0.81) is the tightest pair — two retail-flow brokerages. MSTR is the BTC leg. STRC, Strategy's preferred, correlates 0.58 with the common at 180d but 0.70 over the last 60 days — the "safer" instrument is converging toward the equity. PURRDAT, the HYPE treasury company, tracks HYPE itself at 0.82 on its short history.

Crypto AI

Internally coherent the way all crypto is (0.48 average) — and correlated to AI semiconductors at just 0.28 as a basket. The narrative overlap with Nvidia's supply chain does not show up in daily returns.

Is there lead-lag alpha in any of this?

Contemporaneous correlation is not a trade. So we re-ran the study as a lead-lag test: X's daily return against Y's return on its next traded day, ~280–420 observations per pair, with conditional returns after ±1σ days in X and a split-half stability check — then tested whether Hyperliquid's 24/7 perps already price each effect.

The signals that look like alpha

An 84% hit rate with a 7.5% conditional spread looks like a money printer. It is not — these are overnight gaps. The Asia move happens in the Seoul/Tokyo opening auction, which already reflects the US session; the Monday move in crypto-linked stocks happens in the New York open, which already reflects the weekend. You cannot buy an open before it opens.

The obvious workaround is Hyperliquid's 24/7 equity perps — buy the SK Hynix perp on Friday night, or the MSTR perp on Sunday. We tested whether the perps leave that gap open. They do not. The SKHX and SMSN perps move 0.47–0.49 with NVDA during the same UTC day — the perp absorbs the US move in real time. The MSTR perp tracks weekend BTC at 0.98 — fully priced by Sunday night. In both cases the perp's next-day correlation to the signal is negative (−0.17 to −0.35): chasing the spillover in the perp after the fact is slightly wrong-way.

What survives: a next-day fade after big semiconductor days

Every instrument that trades the semiconductor move same-day shows the same negative next-day sign — classic overreaction and mean reversion, split-half stable, worth roughly 0.9–1.6% per ±1σ event before fees. These legs are correlated, so this is honestly one finding, not five. It is the only pattern in the sweep with a consistent sign, a plausible mechanism, and stability across sample halves; the natural refinement is testing it intraday rather than on daily bars.

And where there is nothing: crypto-internal lead-lag is dead at the daily horizon. BTC does not lead memecoins, DeFi, or alt-L1s (all |t| < 1.5, split-halves flip sign) — a 24/7 market prices its own beta within the day. With ~30 tests run, a couple of t > 2 readings are expected by chance; the fade cluster earns its credibility from sign-consistency across legs, not from any single t-statistic.

Caveats

Appendix: category membership

Non-obvious xyz tickers: DRAM = Roundhill Memory ETF; CBRS = Cerebras Systems; BOT = RoboStrategy; BIRD = NewBird AI; QNT = Quantinuum; PURRDAT = Hyperliquid Strategies (HYPE treasury); SPCX = SpaceX (private-market); STRC = Strategy Series A preferred; XYZ100 = top-100 non-financial US index; SKHX = SK Hynix; SMSN = Samsung Electronics; KIOXIA = Kioxia; LITE = Lumentum; BE = Bloom Energy; BX = Blackstone; USAR = USA Rare Earth; URNM = Sprott Uranium Miners ETF; SMH = VanEck Semiconductor ETF.