Correlations
- Last refresh
- 13 Nov 2026, 06:00 UTC
- Coverage
- 57 crypto + 33 macro references
Macro matrix
Pair laboratory
Interrogate a relationship
Select two assets to calculate price-based pair tests. The correlation cards use the published snapshot; the other tests load public price histories separately.
Research state—
7D correlation—
1M correlation—
1Y correlation—
Persistence score—Correlation-only triage, 0–100
Price-based tests use up to one year of completed daily candles, with at least 180 aligned observations.
Available
The cards above compare Pearson return correlation across 7D, 1M, and 1Y. The persistence score rewards strong co-movement and penalizes differences between windows. Correlation alone does not establish a mean-reverting spread.
Waiting for prices
Tests whether the selected log-price series share a stationary linear combination, with separate checks of their integration order.
Waiting for prices
Fits the selected log prices and measures how the hedge-ratio estimate changes across rolling windows.
Waiting for prices
Standardizes the latest modeled residual against preceding observations and estimates its historical decay.
Waiting for prices
Walks current public order-book depth at the scenario size and adds your fee assumption. Funding and borrow costs are excluded.
Waiting for prices
Evaluates a fixed model on a later chronological holdout, using next-open execution and explicit fee/slippage assumptions.
BTC exposure
Low correlation to BTC
Find tokens with negative or low return correlation to Bitcoin. Sort by the selected window, with the lowest correlation first.
| Token / BTC | 7D | 1M | 1Y | Selected window |
|---|
Daily shortlist
Persistent pair candidates
Ranked by weighted 7D/1M/1Y correlation with a penalty for cross-window instability. High rank means “research next,” not “trade now.”
| Pair | 7D | 1M | 1Y | Range | Persistence | State |
|---|
Regime monitor
Largest short-horizon breaks
Pairs whose 7-day correlation has fallen furthest below the 1-month relationship. This can reflect a catalyst, changing beta, data mismatch, or a temporary dislocation.
Research basis
Why this is useful—and where correlation stops
Correlation identifies duplicated exposure and narrows thousands of possible pairs. Classic pairs research instead forms pairs from historical relative-price behaviour and tests convergence after divergence. Cointegration theory adds the stricter requirement that a linear combination of non-stationary prices is stationary.
Two assets can show high return correlation while their price spread drifts indefinitely. Structural breaks, changing regimes, funding, and execution costs can erase an apparent relationship.
Gatev, Goetzmann & RouwenhorstPairs formed by historical normalized-price distance; convergence is tested after divergence. Engle & GrangerCointegration and error correction provide the basis for a mean-reverting equilibrium relation. Campos, Ericsson & HendryStructural change can distort stationarity and cointegration inference. Bailey et al.Repeated strategy selection creates backtest-overfitting risk; validation must account for multiple trials.
Implementation path
From dependency map to testable signal
- 01 · Live nowCandidate discovery
Multi-horizon Pearson return correlation, aligned-observation counts, persistence ranking, and break detection.
- 02 · Calculated on selectionEquilibrium model
Aligned daily log prices, Engle–Granger cointegration, integration-order checks, and rolling OLS beta estimates.
- 03 · Calculated on selectionSpread state
Residual z-score against prior observations and an AR(1) half-life estimate where stable decay can be estimated.
- 04 · Scenario evaluationCosts and holdout
Current book-depth costs and a chronological holdout with assumed fees/slippage. Funding, borrow, walk-forward testing, and multiple-testing corrections remain outside this calculation.
Proposed publish rule
Only show a directional research signal when cointegration passes, beta is stable, the spread is sufficiently displaced, expected convergence exceeds modeled costs, both legs are liquid, and the model remains valid out of sample. Otherwise publish “candidate,” “monitor,” or “no signal.”
Calculation references: Engle–Granger methodology · Public candles and order books. Unsupported markets and short histories return an explicit unavailable result.
Source
Hourly USDT spot candles from Binance, OKX, MEXC and Gate. Hyperliquid perpetual references at 4h; QQQ at synchronized daily session closes.
Universe
Explicit asset identities; no ticker substitutions. At least 80% aligned window coverage.
Method
Crypto: Pearson correlation of hourly USDT spot log returns from Binance, OKX, MEXC and Gate. Macro and GOLD/IGV pairs: 4-hour perpetual returns. QQQ: synchronized daily session-close returns (minimum 3 / 15 / 150 observations for 7D / 1M / 1Y). Windows require 80% aligned coverage; gaps are excluded.