GLOSSARY DEEP DIVE

Correlation: The Number That Decides If Your Diversification Is Real

Owning ten different stocks feels like diversification, but if those ten stocks tend to rise and fall together in most market conditions, you have not actually reduced your risk very much, no matter how different the company names and logos look on a statement. Correlation is the number that separates genuine, measurable risk reduction from the comforting illusion of it, and most individual investors have never actually looked at it before building a portfolio.

Deep dive10 min readUpdated 2026
In this article

The core principle

Correlation measures how closely two investments move together, expressed as a coefficient between -1 and +1. A correlation of +1 means two assets move in lockstep, always in the same direction, by proportional amounts. A correlation of -1 means they move perfectly opposite each other. A correlation of 0 means their movements are statistically unrelated. In plain terms, the formula is correlation = covariance of the two assets / (standard deviation of asset A × standard deviation of asset B), where covariance captures how much two assets move together and the denominator scales that into a standardized, comparable number that behaves the same way whether you are comparing two stocks, a stock and a bond, or a stock and an entire foreign market index.

In practice, most individual US large-cap stocks carry a correlation to the broad stock market somewhere around 0.7 to 0.9, since they are all exposed to the same macroeconomic forces, interest rates, growth expectations, investor sentiment, even when their businesses have little else in common. Government bonds have historically shown a correlation to stocks much closer to 0, and at various points meaningfully negative, which is precisely why a stock and bond mix tends to smooth out a portfolio's swings far more effectively than simply adding a twentieth or thirtieth individual stock to an already stock-heavy, sector-concentrated portfolio.

Key idea Correlation is not fixed. It tends to rise toward +1 during sharp market crashes, exactly the moment investors most need their assets to move independently, a pattern sometimes called correlation convergence. Diversification tends to work best in calm markets and worst in the crises it was supposed to help with.

It is also worth being clear about what correlation does not tell you. Two assets can be highly correlated in direction, generally rising and falling together, while still differing enormously in magnitude, one moving twice as far as the other on the same news. Correlation captures the pattern of co-movement, not the relative size of the moves, which is a separate concept called beta when measured against a specific benchmark such as the broad market index. A full risk analysis generally needs both numbers together, not correlation in isolation, to accurately describe how two specific assets actually interact inside a real portfolio over time.

How the math works

The formula for a two-asset portfolio's volatility is portfolio variance = w12σ12 + w22σ22 + 2w1w2σ1σ2ρ, where w is each asset's weight in the portfolio, σ is each asset's standard deviation (volatility), and ρ is the correlation between them. The two examples below use identical assets in every respect except correlation, to isolate exactly how much correlation alone changes the outcome.

Example 1: high correlation. Two stocks, each with an 18% annual standard deviation, have a correlation of 0.9, typical of two companies in the same sector. Combined 50/50: variance = 0.52(0.182) + 0.52(0.182) + 2(0.5)(0.5)(0.18)(0.18)(0.9) = 0.0081 + 0.0081 + 0.01458 = 0.03078. The portfolio's standard deviation is √0.03078 ≈ 17.5%, barely below the 18% of either stock alone. Combining two highly correlated assets bought almost no risk reduction at all.

Example 2: low or negative correlation. Same two assets, same 18% volatility each, but now with a correlation of -0.2, more typical of a stock paired with a high-quality government bond during ordinary market conditions. Combined 50/50: variance = 0.0081 + 0.0081 + 2(0.5)(0.5)(0.18)(0.18)(-0.2) = 0.0162 − 0.00324 = 0.01296. The portfolio's standard deviation is √0.01296 ≈ 11.4%, a full 6 percentage points lower than the high-correlation example, even though the two individual assets have exactly the same volatility in both cases and the portfolio's expected return is unaffected by the correlation at all. The entire difference, 17.5% down to 11.4%, comes from a single number: correlation.

Key idea Notice that expected return did not change between the two examples, only risk did. This is the specific, almost free lunch that low correlation offers: a way to reduce portfolio swings without giving up any expected return, which is the entire theoretical justification for diversification in the first place.

How it shows up in real portfolios

During sharp, broad-based market crashes, correlations across asset classes that normally behave independently, developed market stocks, emerging market stocks, corporate credit, even some commodities, have historically tended to rise together toward 1, because the dominant driver in a panic becomes a single factor: forced, indiscriminate selling for liquidity, which overwhelms whatever fundamental differences normally separate those assets. Portfolios that looked well diversified on paper based on calm-market correlation data have repeatedly turned out to be far more correlated than expected exactly when diversification mattered most, a pattern documented across enough separate historical crises that it should be treated as a baseline design assumption, not a surprising footnote, when building a portfolio meant to genuinely hold up under real stress.

A common real-world case involves a technology employee or executive holding a large amount of employer stock alongside a brokerage account concentrated in other technology and growth names, reasoning, understandably but incorrectly, that the portfolio is diversified simply because it holds many different companies across different tickers. In practice, most large technology stocks carry correlations to each other well above 0.6 to 0.8, meaning the portfolio behaves much more like one large, undiversified bet on the sector than the ticker count would suggest. Genuine diversification for that investor requires adding assets, bonds, value-oriented sectors, real assets, that have historically moved with a meaningfully lower correlation to the core holding, not simply adding more technology names.

Institutional portfolio managers formalize this intuition using a full correlation matrix across every asset in a portfolio rather than checking pairs one at a time, since a holding can look fine against any single other position while still adding meaningful, undiversified risk once its relationship to every other holding is considered together. Individual investors rarely need that level of formality, but the underlying discipline, actually checking historical co-movement before assuming a new holding adds diversification, transfers directly and is worth applying to any material addition to a portfolio, whether that addition is a new fund, a new sector, or a large single-stock position.

Actionable breakdown

  • Check correlation before assuming diversification
    • Same sector usually means high correlation
    • Different sectors do not guarantee low correlation
  • Pair asset classes with historically different drivers
    • Stocks and high-quality bonds often qualify
    • Two growth stocks in the same industry rarely do
  • Expect correlations to rise during a crisis
    • Calm-market correlation data understates crash risk
    • Build in a margin of safety for this
  • Question "uncorrelated" claims on complex or costly products
    • Verify the actual historical numbers yourself
    • High fees can erode any diversification benefit
  • Consider the whole portfolio, not just pairs
    • A holding can look fine against one asset
    • Check it against every major position together

Common pitfalls

  • Assuming different company names or ticker symbols mean genuinely different risk exposure, when the underlying correlation between them is close to 1.
  • Relying on correlation figures calculated during calm markets, which routinely understate how correlated assets become during a genuine crisis.
  • Chasing alternative or "uncorrelated" assets marketed on that promise alone, without checking whether the actual historical numbers, net of fees, support the claim.
  • Confusing low correlation with negative expected return, when in fact combining assets with different drivers can reduce risk without sacrificing return at all.
  • Ignoring the difference between correlation and beta, and assuming two assets that move together also move by similar magnitudes when the relationship may be far more lopsided.

See standard deviation for the volatility measure correlation is always calculated alongside, and concentration risk for what tends to happen when correlation gets ignored entirely. Our asset allocation guide and risk guide both build directly on correlation, and bonds covers the asset class most commonly paired with stocks for exactly this purpose in a typical portfolio.

The bottom line

Real diversification comes from deliberately combining assets with genuinely low or negative correlation, verified with actual data, not simply from owning a larger number of things and hoping the names are different enough to matter.

Back to the full glossary