Correlation is a statistic ranging from -1 to +1 that measures the strength and direction of the linear relationship between two variables, such as the returns of two assets.
Correlation
Correlation, usually calculated as the Pearson correlation coefficient and denoted r (or the Greek letter rho, ρ), is a statistical measure of how strongly two variables move together in a straight-line, or linear, pattern. Conceptually, correlation is calculated as covariance(X, Y) divided by the product of the standard deviation of X and the standard deviation of Y. That division is what standardizes the result: covariance on its own can take on almost any value depending on the units and scale of the variables being measured, which makes it hard to interpret at a glance, whereas correlation is always squeezed into the fixed range of -1 to +1.
The scale is straightforward to read once you know the endpoints. A correlation of +1 means two variables move in perfect lockstep — when one rises, the other rises proportionally, every single time. A correlation of -1 means the opposite: the two move in perfect opposite directions. A correlation of 0 means there is no linear relationship at all between the two — knowing the value of one tells you nothing, in a linear sense, about the other. In practice, correlations between real-world assets rarely sit at these extremes; a coefficient around 0.2 to 0.3 is considered low, while anything above 0.7 is generally considered high.
One caveat is essential to keep in mind: correlation only captures linear relationships, so two variables that are related in a more complex, curved, or cyclical way can show a correlation near zero even though they are clearly connected. Correlation is also not a fixed, permanent property of two assets — it is estimated from historical data and can shift meaningfully as market conditions, time periods, and economic regimes change, so it should be reviewed periodically rather than assumed to hold indefinitely.
Example
Consider an investor holding two stocks. Stock A is a more volatile technology company with an annualized standard deviation of 25%. Stock B is a steadier consumer staples company with an annualized standard deviation of 12%. Historical return data shows the correlation between the two is 0.3 — a fairly low positive correlation, meaning they tend to move in the same general direction but not tightly together. Converting that back into covariance: covariance = correlation x standard deviation of A x standard deviation of B = 0.3 x 0.25 x 0.12 = 0.009.
Suppose the investor puts $100,000 to work, allocating $60,000 (60%) to Stock A and $40,000 (40%) to Stock B. A naive weighted average of the two standard deviations would suggest portfolio volatility of 60% x 25% + 40% x 12% = 19.8%. But because the correlation between A and B is only 0.3, their movements only partially overlap. Applying the portfolio variance formula, which incorporates that correlation, produces an actual portfolio standard deviation of roughly 17.06% — noticeably lower than the naive 19.8% estimate.
That 2.74-percentage-point gap is the diversification benefit, and it exists purely because A and B are not highly correlated. If the correlation between the two stocks were 1.0 (perfectly correlated), diversification would provide zero risk reduction and portfolio volatility would simply equal the 19.8% weighted average. The lower — or more negative — the correlation, the larger that diversification benefit becomes.
Practical Application
Correlation sits at the heart of Modern Portfolio Theory and is one of the most frequently cited statistics in portfolio construction. Portfolio managers and financial advisors deliberately seek out asset classes with low or negative correlation to each other — pairing stocks with bonds, or combining equities from different sectors, geographies, and market caps — specifically to reduce overall portfolio volatility without necessarily sacrificing expected return.
In risk management, institutional investors and quantitative trading desks routinely compute correlation matrices across large baskets of assets to gauge how genuinely diversified a portfolio is and to flag positions that carry hidden concentration risk. Banks and insurers also monitor how correlations shift under stress scenarios, since assets that appear loosely related in calm markets frequently move together when systemic risk spikes.
For individual investors, understanding correlation helps avoid a common trap: holding several funds or stocks that look different on the surface but are, in fact, highly correlated — such as multiple technology-heavy funds — and mistakenly believing a portfolio is well diversified when it is really concentrated in one underlying risk factor.
Common Mistakes
The most common mistake is confusing correlation with causation. Two variables can show a strong correlation without one causing the other — the relationship might be driven by a shared third factor, or it might simply be coincidental over the period measured.
A second frequent error is treating correlation as a fixed, permanent number. Correlations are estimated from historical data and shift as market conditions change. Critically, many asset pairs that show low correlation during calm markets tend to converge toward 1 during crises, as panic selling drags nearly all risk assets down together — a phenomenon often summarized as correlations going to 1 in a crisis, and it can undermine diversification exactly when investors need it most.
Some investors assume a correlation of 0 means two variables are completely unrelated, but it only means there is no linear relationship. A more complex, non-linear connection between the two can still exist and simply won't show up as a meaningful correlation coefficient, which is why analysts often pair correlation analysis with a scatter plot or other statistical checks.
Finally, it's easy to overlook how much the length of the data sample affects a correlation estimate. Correlations calculated from very short historical windows can be distorted by short-term noise, so it's generally more reliable to use a longer observation period and to refresh the calculation periodically rather than relying on a single, dated figure.
Comparison
Dimension
Correlation
Covariance
Definition
A standardized measure of the strength and direction of the linear relationship between two variables
An unstandardized measure of whether two variables move in the same or opposite direction
Range / calculation
Covariance / (std dev of X x std dev of Y); always fixed between -1 and +1
Average of the product of each variable's deviation from its own mean; no fixed range
What it indicates
+1 = perfect positive relationship, -1 = perfect negative, 0 = no linear relationship
Positive = move together, negative = move oppositely, but magnitude is hard to interpret alone
Typical use case
Comparing the strength of relationships across different asset pairs; assessing diversification
An intermediate input for calculating correlation and portfolio variance (risk) formulas
Key limitation
Captures only linear relationships and can shift meaningfully over time or across market regimes
Its magnitude depends on the units and scale of the variables, so it can't be compared directly across pairs
Does a correlation of 0 mean two assets are completely unrelated?
Not necessarily. A correlation of 0 only means there is no linear relationship between the two variables — you can't describe how they move together using a straight line. If the two are related in a more complex, non-linear, or cyclical way, correlation can still come out near zero even though a real relationship exists. To check for that possibility, analysts typically look at a scatter plot of the data alongside the correlation figure rather than relying on the number alone.
Why do portfolio managers look for low-correlation assets?
Because a lower correlation means two assets' returns don't move in lockstep — when one falls, the other may hold steady or even rise, partially offsetting the loss. This offsetting effect means a portfolio's actual volatility ends up lower than what you'd get by simply taking a weighted average of each asset's individual volatility. That risk-reduction mechanism is the central insight behind diversification in Modern Portfolio Theory.
Can correlation between two assets change over time?
Yes, and sometimes significantly. Correlation is calculated from historical return data, so it shifts as macroeconomic conditions, industry dynamics, and investor sentiment evolve. A particularly important pattern is that many asset pairs with low correlation in normal markets tend to move much more closely together during financial crises, as widespread selling pressure hits nearly everything at once — a limitation investors should account for when designing a diversification strategy.
Is correlation or covariance better for comparing different asset pairs?
Correlation is generally the better tool for comparison because it's standardized to a fixed -1 to +1 range regardless of the assets' units or volatility levels, so you can directly compare the strength of the relationship across different pairs. Covariance's magnitude depends on how volatile each variable is, so two pairs that are equally strongly related can show wildly different covariance values, making cross-comparison misleading. Covariance remains essential, though, as the building block used to calculate both correlation and portfolio risk.
Is correlation only used for stocks?
No — correlation is a general-purpose statistical tool that can be applied to any two sets of paired numerical data. Beyond stock returns, it's commonly used to study relationships between bonds, currencies, commodities, real estate, and even macroeconomic indicators like interest rates and inflation. In multi-asset portfolio construction, investors routinely examine correlations across stocks, bonds, and alternative investments together to build a more thoroughly diversified allocation.