Correlation: when everything falls together

Two positions that move together are one position wearing two names. That is the entire reason correlation matters to a trader: it decides whether your portfolio is genuinely spread across different bets or just looks busy. And unlike position size, which you control exactly, correlation is a property of the market — measured imperfectly, drifting constantly, and at its least reliable at exactly the moment you need it most.

You do not need the formula to use the concept well. You need three ideas: what the number means, why it refuses to sit still, and what happens to it in a crisis.

Correlation without the math

Correlation runs from +1 to −1. At +1, two assets move in lockstep; at −1, they mirror each other; at 0, they are strangers. Real pairs live in between, and the number is always computed over some lookback window of past returns — which means it is a description of the recent past, not a law of nature. The same two assets can show +0.8 over one month and +0.2 over a year. Neither number is wrong. Both are photographs of different moments.

• +0.8 and above: effectively the same trade for risk purposes. Size the pair as one position.
• +0.3 to +0.7: related. Diversification helps, but less than the ticker count suggests.
• Around 0: genuinely independent — the valuable stuff, and rarer than it looks.
• Negative: one tends to cushion the other. Precious when real, and almost never stable enough to lean on.

Make it concrete with sizing. Suppose you risk 1% on one coin and 1% on a second coin that tracks the first at +0.9. You do not hold two independent 1% bets; you hold roughly 1.9% of exposure to a single idea. Scale it up: five crypto positions at 1% each, all correlated at +0.8 or better, is most of a single 5% bet — five times the per-trade risk you thought you signed up for. Nothing on the order ticket warned you, because order tickets price positions one at a time. Correlation is the bookkeeping the ticket leaves out.

Correlations are regime-dependent

The relationships between markets are not fixed properties; they are products of who is trading them and what story dominates. Crypto and tech stocks have gone through stretches of moving almost tick for tick — typically when the same liquidity conditions drive both — and stretches of near-total indifference to each other. Stocks and bonds spent decades as a textbook negative-correlation pair, then spent other stretches falling together whenever inflation was the dominant fear. The driver changes, and the correlation changes with it.

This is why a correlation matrix from last quarter is a museum piece. The honest posture is to treat any measured correlation as a tendency with an expiration date, and to look at several lookback windows rather than trusting one number to summarize a relationship.

Crisis convergence

Now the part that matters most for survival. In a genuine liquidation event — a fast, leveraged, everyone-out-of-the-pool decline — correlations converge toward one. Assets that ignored each other for years suddenly fall in formation. The mechanism is not mysterious: when leveraged players get margin calls, they do not sell what they want to sell, they sell what they can sell. Quality gets liquidated alongside junk because quality has a bid. Gold drops with stocks with crypto, and the safe-haven currency people are fleeing into — historically, often the dollar — is the one thing rising.

The cruel implication: diversification works best on ordinary days and worst on the days you bought it for. It is seatbelt-grade protection, not force-field-grade. The force field, such as it exists, is position sizing — which keeps working when correlations do not.

The honest cross-asset map

With the instability warning firmly attached, here are the tendencies traders actually observe across the four classes — tendencies, not laws:

• Crypto and stocks: episodically high, especially during risk-off periods and shared liquidity squeezes; closer to zero in quiet stretches. Treat them as cousins, not strangers.
• Gold and stocks: often low or negative, which is gold’s whole reputation — except in liquidity crunches, when gold can fall with everything as traders raise cash, before frequently recovering earlier than the rest.
• The dollar and almost everything: dollar strength tends to pressure crypto, metals and risk-sensitive currencies all at once. It is the closest thing to a hidden common factor across the board.
• Within a class: crypto-to-crypto correlations are notoriously high, and major forex pairs cluster around the direction of the dollar. Intra-class diversification is the weakest kind.

What to do about it

You cannot control correlation, but you can stop it from ambushing you. The defenses are all sizing and bookkeeping, not prediction.

• Group your positions by risk driver and treat each cluster as one position when allocating risk — a watchlist per cluster makes this visible at a glance.
• Cap the total risk in any single cluster, exactly as you would cap a single trade.
• Check relationships over multiple windows occasionally; when a usually-independent pair starts moving together, your real exposure just grew without you trading.
• Let diversification earn its keep on average days, and let small position sizes carry you through the convergent ones.

Key takeaways

• Correlation measures co-movement on a scale from −1 to +1, always computed over a past window — it describes, it does not promise.
• Relationships between asset classes are regime-dependent and flip without notice; treat any measured correlation as perishable.
• In liquidation events, correlations converge toward one because forced sellers sell whatever has a bid.
• Crypto, stocks, gold and the dollar have observable tendencies, all of them unstable — the dollar is the closest thing to a common factor.
• Size positions so that the everything-falls-together day is survivable; that is the one correlation forecast you can rely on.