Cointegration is not correlation. Two assets can be highly correlated today and lose that correlation tomorrow — correlation is unstable. Cointegration is a deeper statistical property: it says the spread between two assets is stationary, meaning it reverts to a mean. When you see SPY and GLD are cointegrated, it means their relative pricing has an anchor — deviations are temporary.
The Engle-Granger test formalizes this. A p-value below 0.05 means the null hypothesis of "no cointegration" is rejected — the relationship is statistically real. Between 0.05 and 0.10, it's under strain. Above 0.10, it's gone.
SPY / GLD — Risk vs. haven equilibrium. When this breaks, the traditional equity-gold hedge is no longer functioning. Investors who rely on gold as a portfolio diversifier need to know this immediately.
QQQ / SPY — Tech premium stability. Is tech still leading broad equity, or has the AI premium become its own animal? A break here means tech is trading on its own narrative, disconnected from the broader market.
BTC-USD / QQQ — The canary pair. Crypto and tech have been linked because both attract the same speculative capital. When this breaks, speculative liquidity is either draining or finding new channels. This pair tends to break first.
GLD / UUP — The gold-dollar inverse. One of the most reliable macro anchors for decades. When it breaks, central bank policy regimes are shifting — think de-dollarization, reserve diversification, or a fundamental change in how the dollar functions.
TLT / SPY — Equity-bond correlation. This defines the 60/40 portfolio. When bonds and equities become positively correlated (both falling together), traditional portfolio construction fails. This happened in 2022 and the regime shift lasted over a year.
VEU / SPY — US vs. rest-of-world. When this breaks, capital is structurally rotating between geographies — not just tactically shifting for a quarter, but repricing where growth lives.
Cointegration breaks don't happen in isolation. There's a sequence — a domino chain. The model tracks five dominoes in order: Crypto→Tech → Tech→Broad → Equity→Haven → Bond→Equity → Gold→Dollar.
The sequence matters because it maps to how stress propagates. First, speculative capital exits (crypto decouples from tech). Then the tech premium compresses. Then broad equities break their relationship with havens. Then the bond-equity correlation flips. Finally, the deepest anchor — gold-dollar — breaks.
When you see dominoes 1 and 2 falling, you have a window to act before 3, 4, and 5 follow. That's the operational value of cascade tracking.
Half-life tells you how many days the spread takes to revert halfway to its mean. Under 15 days means the relationship snaps back fast — deviations are trading opportunities. Over 30 days means the relationship is loose — deviations could persist.
Hurst exponent below 0.5 confirms mean-reversion. Above 0.5 signals trending behavior — the spread is walking away rather than reverting. A Hurst reading of 0.6+ on a previously mean-reverting pair is a structural warning.
Variance ratio near 1.0 means random walk (no predictability). Well below 1.0 confirms mean-reversion. The variance ratio and Hurst should agree — when they diverge, the statistical picture is murky and caution is warranted.
The Cointegration Monitor runs daily on a rolling 60-day window. Status changes (LOCKED → STRETCHING → BREAKING) are flagged in the Temperature Daily Reading and Weekly Signal. When a pair transitions to BREAKING, it stays flagged until cointegration re-establishes — which can take weeks to months.
Live Monitor →Money Temperature →
