Correlation: How Asset Returns Move Together and Why It Changes

What is Correlation?

Correlation measures the strength and direction of the linear relationship between two return series.

For investment returns, the usual Pearson correlation coefficient ranges from -1 to +1:

1+1.0  -> perfectly positive linear relationship
2 0.0  -> no linear relationship in the sample
3-1.0  -> perfectly negative linear relationship

A correlation of +0.80 means the two return series tended to move strongly in the same direction over the measured sample. A correlation of -0.40 means they tended to move in opposite directions. A value close to zero means little linear co-movement was observed.

The word observed matters. Correlation is an estimate made from a particular set of data, not a permanent property stamped onto two assets.

A simple investment example

Imagine two assets with monthly returns that often rise and fall together. Their correlation might be high and positive.

Now imagine pairing an equity index with an asset whose returns often respond differently to economic shocks. The measured correlation might be lower, creating more room for Diversification.

Suppose a portfolio holds two assets with equal volatility:

1Asset A volatility: 15%
2Asset B volatility: 15%

If their correlation is +1.0, combining them does not reduce risk through co-movement. If their correlation is 0.0, the combination can have lower volatility than either asset alone. If their correlation is negative, the reduction can be larger.

That is why portfolio construction cares about relationships between holdings, not just each holding's standalone Volatility.

Correlation is standardized covariance, not causation

Correlation is closely related to Covariance:

1Correlation(A,B)
2=
3Covariance(A,B)
4-----------------------------
5Standard Deviation(A) × Standard Deviation(B)

Dividing covariance by the two standard deviations standardizes the relationship. That is what puts correlation on the fixed -1 to +1 scale.

This also creates an important analytical boundary:

Correlation is standardized covariance, not causation.

Two assets can be correlated because they share economic drivers, because one indirectly influences the other, because both respond to a third variable, or simply because the chosen sample produced an apparent relationship.

A high correlation alone does not identify the economic mechanism.

Zero correlation does not mean independence

A correlation near zero says little linear co-movement was measured. It does not prove the two variables are statistically independent.

Nonlinear relationships can exist even when linear correlation is zero.

For example, an asset could react strongly to very large market moves in either direction while showing little linear relationship during ordinary periods. A single correlation coefficient could miss that shape.

Investors should therefore avoid translating:

1correlation ≈ 0

into:

1these assets have nothing to do with each other

The first statement is about a specific statistic. The second is a much stronger claim.

The sample changes the answer

A historical correlation estimate depends on choices including:

• daily, weekly, or monthly returns;
• the start and end dates;
• the length of the lookback window;
• how missing observations are handled;
• whether prices are synchronized to the same market close;
• currency treatment; and
• whether the return series include distributions and corporate actions consistently.

A one-year daily correlation can differ sharply from a ten-year monthly correlation.

Neither number is automatically wrong. They may describe different horizons and regimes.

When two data providers publish different correlations, the first question should be whether they used the same inputs and methodology.

Correlations can rise during market stress

Diversification can look strongest during calm periods and weaker precisely when investors need it most.

During a broad liquidity shock, many risky assets can be sold at the same time. Shared exposure to growth, funding conditions, leverage, or investor risk appetite can pull formerly distinct return streams together.

That means a low historical correlation is not a promise that two assets will remain loosely related during a crisis.

Historical correlation can change during stress.

This does not make correlation useless. It means portfolio analysis should test more than one window and consider how the underlying economic exposures might behave in adverse regimes.

Holding more securities is not enough

Suppose an investor owns 25 technology stocks. The position count is large, but the stocks may share exposure to the same industry, valuation factor, interest-rate sensitivity, and economic cycle.

Their pairwise correlations can remain high.

A second investor might hold fewer securities but spread exposure across genuinely different economic drivers.

The second portfolio can be better diversified even with fewer line items.

Correlation helps explain why Diversification is about independent sources of risk rather than collecting tickers.

Correlation enters portfolio variance directly

For a two-asset portfolio, risk depends on both standalone volatilities and the correlation between returns:

1Portfolio Variance
2= wA² σA²
3+ wB² σB²
4+ 2 wA wB σA σB ρAB

where ρAB is the correlation between assets A and B.

The final term is why two portfolios with identical weights and identical standalone asset volatilities can still have different total risk.

Lower correlation reduces that cross term, all else equal.

For larger portfolios, the same principle expands into a covariance matrix. The number of relationships grows quickly, making estimated co-movement a central input to Portfolio Variance and mean-variance optimization.

Negative correlation is not automatically better

A negative correlation can be valuable for diversification, but investors still need to evaluate expected return, liquidity, costs, leverage, taxes, and the economic reason for the relationship.

An asset that reliably loses money can reduce correlation without improving the portfolio's objective.

Likewise, a hedge can have a negative expected carry but still be useful because it performs during specific stress scenarios.

Correlation is one property of a portfolio component, not a complete investment decision.

Correlation matrices contain estimation risk

An optimizer may use hundreds or thousands of pairwise correlations.

Those estimates are noisy. Small changes in the inputs can materially alter an Efficient Frontier or the weights of a Minimum Variance Portfolio.

Common approaches to reduce instability include:

• longer estimation windows;
• shrinkage toward more stable structures;
• factor models;
• stress scenarios;
• weight constraints; and
• explicit turnover or transaction-cost limits.

Each approach trades some responsiveness for robustness.

Correlation is not a tail-loss measure

A pairwise correlation does not tell an investor:

• how much either asset can lose;
• the largest observed drawdown;
• the loss threshold at a chosen confidence level;
• the average loss in the worst tail; or
• how quickly liquidity might disappear.

Those questions require other tools such as Maximum Drawdown, Value at Risk, and Expected Shortfall.

Correlation can influence each of those outcomes, but it is not a substitute for them.

How investors should use correlation

A useful correlation review asks:

1. Which return series are being compared?
2. What frequency and lookback window were used?
3. Are the prices aligned in time and currency?
4. Is the relationship economically plausible or merely statistical?
5. How did the correlation behave in stressed periods?
6. Is the relationship stable across subperiods?
7. Are several holdings driven by the same hidden factor?
8. What happens to portfolio risk if the correlation rises?

Grizzly Bulls' Models can be reviewed with these questions in mind without treating this page as a live correlation estimator. The Macroeconomic Conditions Index provides separate regime context that can help frame why relationships might change, but it is not an input to the historical correlation calculation described here.

Sources and further reading

• CFA Institute, 2026, Portfolio Risk and Return: Part I: https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/portfolio-risk-return-part-1
• CFA Institute, 2026, Portfolio Risk and Return: Part II: https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/portfolio-risk-return-part-2
• CFA Institute, 2026, Portfolio Mathematics: https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/portfolio-mathematics

Explore more topics in the Financial Research Encyclopedia.