Correlation, Covariance & the Correlation Matrix

Definition

Correlation measures how closely two investments move together, on a scale from −1 to +1. At +1 they move in perfect lockstep: when one rises 2%, the other rises proportionally every time. At −1 they move in perfect opposition. At 0 there is no linear relationship at all — knowing what one did today tells you nothing about the other. Real-world pairings land somewhere in between: two large-cap U.S. stock funds might sit around +0.9, a stock fund and a high-quality bond fund might hover near 0, and truly negative pairings are rare and usually temporary.

Correlation has an unscaled cousin called covariance. Covariance also captures whether two return streams move together, but it comes out in awkward squared-percentage units whose size depends on how volatile each asset is — so a covariance of 0.002 is meaningless on its own. Correlation fixes that by dividing covariance by each asset's standard deviation, which rescales the number onto the clean −1-to-+1 range:

correlation = covariance(A, B) / (σA × σB)

  covariance(A, B) = average of (A's deviation × B's deviation), period by period
  σA               = standard deviation of asset A's returns
  σB               = standard deviation of asset B's returns

In plain English: for each period, ask how far asset A landed from its own average and how far asset B landed from its own average, then multiply those two deviations. When both assets beat their averages together (or miss together), the product is positive; when one is up while the other is down, it is negative. Average those products across all periods and you have covariance — a raw "do they move together" score. Dividing by the two standard deviations strips out how *big* each asset's swings are, leaving only how *synchronized* they are. That is why correlation is comparable across any pair of assets while covariance is not.

If you have read the companion article on beta, this formula should look familiar: beta is built from the same covariance idea, measured against the market and left partially unscaled. Square a fund's correlation with the market and you get R-squared — the share of its movement the market explains.

Why It Matters

Correlation is the engine that makes diversification actually reduce risk. The introductory article shows *that* combining imperfectly correlated holdings smooths a portfolio; here is *why*, and by how much.

A portfolio's volatility is not the weighted average of its holdings' volatilities — it is almost always less, and correlation is the reason. When two holdings are not perfectly synchronized, some of their swings arrive at different times and partially cancel. The math only allows the portfolio's standard deviation to equal the weighted average in the extreme case of correlation exactly +1. Every step below +1 buys you a volatility discount without touching expected return, which is why diversification gets called the only free lunch in investing.

For income investors this matters in a specific way: the goal is usually a steady payout stream you can live on, and the biggest threat to staying invested is a portfolio that drops hard all at once. A collection of dividend funds that are all +0.9 correlated with each other will fall as one in a bad market, no matter how many tickers it contains. Correlation is the number that reveals this *before* the drawdown does — and the correlation matrix (below) is the tool that shows it for the whole portfolio at a glance.

Correlation also quietly powers the metrics you already use: the Sharpe ratio's denominator is portfolio standard deviation, which depends on every pairwise correlation inside the portfolio. Lower the correlations and the same return earns a better risk-adjusted score.

Example

Take two funds, each with an annualized standard deviation of 15%, held 50/50, with a correlation of +0.3 — roughly what a stock fund and a diversified alternative-income fund might show. All numbers are illustrative. Portfolio variance for two assets is:

σp² = wA²σA² + wB²σB² + 2 × wA × wB × ρ × σA × σB

    = (0.5² × 15²) + (0.5² × 15²) + (2 × 0.5 × 0.5 × 0.3 × 15 × 15)
    = 56.25 + 56.25 + 33.75
    = 146.25  (in %² units)

σp  = √146.25 ≈ 12.1%

The weighted average of two 15% volatilities is, of course, 15%. But the actual portfolio lands at roughly 12.1% — about a fifth of the risk canceled out purely by imperfect correlation, with no change to either holding. Rerun the same math at other correlations and the pattern is clear: at ρ = +1 the portfolio sits at the full 15% (no benefit at all); at ρ = 0 it drops to about 10.6%; negative correlations shrink it further still. The lower the correlation, the deeper the discount.

Now the income-investor version. Suppose you hold SCHD, a dividend-growth ETF, and you add QQQI, a covered-call fund on the Nasdaq-100. Both are equity funds drawing on large-cap U.S. stocks, so their correlation is high — the combination smooths only a little. Add BND, a broad investment-grade bond ETF, and the story changes: stock-bond correlations are typically far lower, so even a modest bond sleeve pulls the portfolio's volatility meaningfully below the weighted average. Same three tickers on a statement; very different diversification work being done by each pairing.

Reading the Correlation Matrix

A single correlation describes one pair. A correlation matrix describes every pair in your portfolio at once — a grid with each holding along both the rows and the columns, where the cell at row A, column B holds the correlation between A and B. Three features make it quick to read:

• The diagonal is always 1.00, because every holding is perfectly correlated with itself. It is a landmark, not information.
• The grid is symmetric — the correlation of A with B equals the correlation of B with A, so you only need to read one triangle of the grid.
• The color of the off-diagonal cells is the story. A matrix full of high values (+0.8 and up) means the portfolio moves as one block; scattered low and moderate values mean the holdings genuinely offset each other.

Here is an illustrative matrix for a five-fund income portfolio (figures illustrative, based on typical long-run relationships; real values shift over time):

Read it like this. The top-left block — S&P fund, dividend ETF, covered-call fund — is a wall of +0.85 to +0.92. Those three are, statistically, close to one bet in three different wrappers: same large-cap U.S. equity engine under different hoods. This is the correlation matrix's version of the ETF overlap problem covered in diversification — overlap in *holdings* shows up here as high correlation in *returns*, and the matrix catches it even when two funds share few actual positions but respond to the same forces. The REIT fund, at roughly +0.7 to the equity block, adds some independence but less than its "real estate" label suggests. The bond ETF is the outlier doing the heavy lifting: at +0.12 to +0.20 against the equity funds, it is the one holding whose swings genuinely arrive at different times.

Rule of thumb: scan the off-diagonal cells of your matrix. If nearly everything reads +0.8 or higher, your fund count is cosmetic — the portfolio will behave like a single position in a selloff, and its income stream depends on one engine.

When Correlation Misleads

Correlation is powerful, but it comes with three honest caveats that matter most exactly when you are counting on it.

Correlations are unstable — and they rise in crashes. The numbers in any matrix are historical averages over some window, and they drift as regimes change. Worse, they are least reliable at the worst moment: in a panic, investors sell nearly everything at once, and pairings that measured +0.5 in calm markets can spike toward +0.9. Traders summarize this as "in a crisis, correlations go to 1." Diversification built only on moderately correlated equity funds tends to evaporate precisely when you need it, which is why max drawdown is worth checking alongside any correlation figure.

The stock-bond relationship shifts regimes. For much of the 2000s and 2010s, U.S. stocks and Treasury-grade bonds were near zero or negatively correlated — bonds rallied when stocks fell, making them a superb shock absorber. In 2022, with inflation driving both markets, stocks and bonds fell together and the correlation flipped firmly positive. Neither regime is permanent. Treat any stock-bond correlation as a snapshot, not a law of nature.

Low correlation is not low risk. Correlation says nothing about how large or dangerous a holding's swings are on its own — that is what standard deviation measures. A speculative asset can be nearly uncorrelated with your equity funds and still be capable of losing half its value. Correlation describes the *relationships*; it is silent about each piece's own risk and quality.

Common Mistakes

• Judging diversification by ticker count instead of correlation. Five income funds at +0.9 to each other diversify barely better than one. The matrix, not the number of holdings, tells you how many independent bets you really own.
• Confusing covariance with correlation. Covariance is unscaled and not comparable across pairs; a bigger covariance may just mean more volatile assets, not a tighter relationship. Compare correlations, not covariances.
• Treating a historical matrix as permanent. Correlations measured over a calm window understate crisis behavior, and stock-bond correlation has flipped sign across regimes. Re-check the matrix periodically rather than filing it away.
• Assuming a low-correlation asset is "safe." Uncorrelated is not the same as stable. Check the holding's own volatility and drawdown history before crediting it as a diversifier.
• Expecting negative correlation from normal assets. Almost everything with a positive expected return is positively correlated with stocks over long horizons. The realistic goal is *low* positive correlation, not −1.
• Ignoring what drives the correlation. Two funds can correlate at +0.9 because they hold the same stocks, or because they respond to the same interest-rate and growth forces. The fix differs — overlap analysis diagnoses the first; changing asset allocation addresses the second.

FAQ

What is a good correlation for diversification?

There is no magic threshold, but rough bands help. Pairings above +0.8 provide little diversification — the holdings are effectively the same bet. Between roughly +0.3 and +0.7 you get real but partial smoothing. Below +0.3 — typical historically for high-quality bonds versus stocks — a holding does serious diversification work. Anything negative is a genuine shock absorber, but sustained negative correlations are rare among assets with positive expected returns. The practical goal is a matrix with plenty of cells below about +0.6, not a hunt for −1.

What does a correlation of 1 mean?

A correlation of exactly +1 means two holdings move in perfect lockstep — every period, their returns land in the same direction, in strict proportion. Owning both provides zero diversification benefit: portfolio volatility equals the plain weighted average of the two, the only case where that is true. Note that +1 does not mean *identical* returns — one asset can be a leveraged, larger-swinging version of the other and still correlate at +1.

Why did my "diversified" funds all fall together?

Two likely reasons. First, overlap: many dividend and covered-call ETFs draw from the same large-cap U.S. universe, so their correlations run +0.85 or higher — different labels, one engine, as the matrix above shows. Second, crisis behavior: correlations measured in calm markets rise sharply in selloffs as investors dump nearly everything at once, so even moderately correlated funds converge exactly when it hurts. A look-through analysis of your funds' underlying holdings usually reveals how much of your diversification was cosmetic.

What is the difference between correlation and covariance?

They capture the same underlying idea — whether two return streams move together — at different scales. Covariance is the raw average of paired deviations, expressed in squared-percentage units that depend on each asset's volatility, so it cannot be compared across pairs. Correlation divides covariance by both assets' standard deviations, rescaling it to a universal −1-to-+1 range. Covariance is the ingredient used inside portfolio math and beta; correlation is the human-readable version you should actually compare.

Is negative correlation better than zero correlation?

For pure risk reduction, yes — a negatively correlated holding actively offsets losses rather than merely sitting out the storm, so it shrinks portfolio volatility more per dollar. The catch is cost: assets that stay reliably negatively correlated with stocks often bring lower expected returns or ongoing drag, and the relationship can flip — as stock-bond correlation did in 2022. For most income portfolios, cheap low-positive correlation beats expensive negative correlation.

Does low correlation mean low risk?

No. Correlation describes a holding's *relationship* to other holdings, not its own risk. An asset can be nearly uncorrelated with your equity funds and still carry violent swings, deep drawdowns, or an unreliable distribution. Always read correlation alongside the holding's own standard deviation and max drawdown: the matrix tells you how the pieces fit together, while those metrics tell you whether each piece is safe to hold at all.