Cross-Asset Correlations: Why Relationships Change
Correlation describes the strength and direction of a statistical relationship between two return series over a defined sample. Cross-asset correlations are not permanent economic laws: stocks and bonds can move together in one regime and in opposite directions in another. Diversification based only on an old average can disappoint when market conditions change.
How correlation is measured
Pearson correlation is covariance of two return series divided by the product of their standard deviations. It ranges from −1 to +1 for defined samples, with zero meaning no linear association rather than absolute independence. The estimate depends on return frequency, sample length, price versus total returns and currency. Comparing a daily gold series with monthly stock returns without matching dates produces an invalid analysis.
Worked example: a strong positive relationship
Suppose two fictional assets have sample covariance of 0.00015 using decimal daily returns and standard deviations of 0.015 and 0.012. Their correlation is 0.00015/(0.015×0.012)≈0.83. This indicates strong positive linear co-movement over that sample, but does not mean they return the same percentage or will keep moving together. A more volatile asset can have far larger price swings even with high correlation.
Why stock-bond correlation changes
When growth fears dominate, government bonds may rise as shares fall, producing negative equity–bond return correlation. When inflation and interest-rate surprises dominate, bond prices and equity valuations can both fall, producing positive correlation. Policy credibility, duration supply and valuation levels also matter. Thus the familiar rule that bonds always protect against stock losses is not reliable across every regime.
Gold and US-dollar relationships
Gold is denominated internationally in dollars, leading to a sometimes negative association between gold and DXY. But real yields, jewellery demand, investment flows and risk aversion can cause gold and the dollar to rise together. A reported correlation without time period, price definition and currency denominator does not support a universal trading rule. Gold’s role can change as macroeconomic conditions evolve.
Rolling correlations and overlapping windows
A rolling 60-day correlation recalculates the statistic for each adjacent 60-day return window. It can reveal changing regimes but is noisy and heavily overlapping: consecutive estimates share most of their observations. A sudden spike may arise from one extreme day, and the estimate’s significance depends on sample size. Compare multiple windows and examine underlying scatterplots where possible.
Diversification and portfolio variance
For two assets, portfolio variance includes each asset’s own variance plus twice the product of weights, standard deviations and correlation. Reducing positive correlation can reduce variance, holding all other inputs fixed. But correlations often increase among risky assets during stress, precisely when diversification is needed most. Tail dependence and nonlinear losses may not be adequately captured by ordinary Pearson correlation.
Causation versus correlation
Two assets can co-move because both respond to a third force such as interest rates. Their observed relationship need not mean that one directly drives the other. A pair of assets can also exhibit a high correlation due to common time trends rather than economically linked returns. Work with returns, matched observations and plausible mechanisms, and use a disciplined out-of-sample framework for predictive claims.
A robust cross-asset dashboard
Track asset-return correlation alongside inflation surprises, yield changes, growth data and risk sentiment. State currency and return conventions. Use rolling comparisons, stress scenarios and historical episodes without assuming that the latest regime lasts forever. Correlation is one input into risk analysis, not a substitute for position sizing or cash-flow research.
For the related market mechanism, see risk-on and risk-off correlations.
How cross-asset correlations can change with the regime
These numbers are hypothetical correlation coefficients between −1 and +1 over unspecified observation windows. They are teaching examples, not empirical estimates for any historical period.
| Illustrative historical window | Stocks vs high-quality bonds | Stocks vs credit | Interpretation |
|---|---|---|---|
| Disinflationary growth slowdown | −0.30 | +0.60 | Bonds may offset some stock risk |
| Inflation shock | +0.50 | +0.70 | Both stock and bond values can decline |
| Liquidity shock | Relationships can jump or change sign | Relationships can converge | Forced selling and market mechanics dominate |
Correlation is not causation and does not remain fixed when interest-rate, inflation and liquidity conditions change. In a portfolio, an apparent hedge can stop behaving defensively when several assets share exposure to rising discount rates. Estimates depend on return frequency, currency, hedging, sample window and outlier treatment. Compare rolling results with economically plausible mechanisms rather than simply treating the most recently calculated coefficient as a structural law.
Frequently asked questions
Is zero correlation the same as independence?
No. Nonlinear dependence may remain.
Do bonds always rise when stocks fall?
No. Inflation and policy shocks can cause both to decline.
Can gold and the dollar rise together?
Yes. Macro and risk-demand regimes can support both.
Sources and related education
Related learning: How Interest Rates and Inflation Affect Financial Markets.