Correlation measures how closely the returns of two assets move together. It is expressed as a single number between -1 and +1:
Most real pairs of stocks land somewhere in between — often mildly positive, because they share exposure to the same overall market.
Dividing covariance by the two standard deviations rescales it into a fixed -1 to +1 range, so pairs of any volatility can be compared directly.
Highly correlated (≈ 0.9). Two large-cap tech names like Apple and Microsoft often move at around 0.9. Holding both feels like diversification, but at that level it is closer to one bet in two wrappers — they rise and fall together and rarely cushion one another.
Roughly independent (≈ 0.03). A tech stock and an exchange operator such as CME can sit near zero. Their day-to-day moves barely relate, so pairing them genuinely spreads risk.
Inversely correlated (≈ -0.4). An asset like gold (or a long-Treasury fund) often carries a negative correlation to equities. When stocks sell off it tends to hold up or rise, which is what smooths a portfolio’s ride.
Correlation is the honest test of diversification. Owning twenty stocks that all sit at 0.9 to each other is not a diversified portfolio — it is one large position wearing twenty name tags. Real diversification comes from combining holdings whose correlations are low or negative, so a drawdown in one is not mirrored across the rest.
A correlation matrix lays every pair out at once. Read it like a grid: each cell is the correlation of the ticker in its row with the ticker in its column. The matrix is symmetric — the value for A↔B equals B↔A — and the diagonal is always 1, because every asset is perfectly correlated with itself and therefore carries no information. Scan for the hot cells: clusters of high correlation are concentration risk hiding in plain sight.
Correlation matters most when you are building or stress-testing a portfolio. Two positions that look different on paper but move at 0.95 add almost nothing to resilience, while a genuinely uncorrelated or inversely correlated holding can cut overall volatility and shrink drawdowns without giving up much expected return.
Correlation is backward-looking and unstable: it is estimated over a chosen window and shifts as that window moves. Worse, correlations tend to rise toward 1 in a crisis — exactly when diversification is supposed to help, assets that normally drift apart can crash together. It also captures only linearco-movement and says nothing about which asset causes the other to move.