By Algovestiq Research Team
Correlation in Investing
Correlation measures how two assets' returns move relative to each other — it is the mathematical engine behind portfolio diversification, and understanding how it changes across market regimes separates investors who are genuinely diversified from those who only appear to be.
What Correlation Measures
Correlation (ρ) is a dimensionless measure from -1.0 to +1.0 that quantifies the linear relationship between two return series. A correlation of +1.0 means the assets move together perfectly — when one rises 5%, the other always rises 5%. A correlation of -1.0 means they move in perfectly opposite directions — a theoretically perfect hedge. A correlation of 0 means no linear relationship — the assets move independently of each other.
ρ(A,B) = Covariance(A,B) / (StdDev(A) × StdDev(B))Portfolio variance = w₁²σ₁² + w₂²σ₂² + 2w₁w₂σ₁σ₂ρ₁₂. The correlation term (the last component) directly determines how much portfolio volatility is reduced through combining assets. When ρ = 1.0, no reduction occurs. When ρ = 0, portfolio variance equals the weighted sum of individual variances — genuine risk reduction. When ρ = -1.0, variance can be eliminated entirely — a perfectly hedged portfolio. This formula explains why diversification is not about how many positions you hold but about the correlation structure between them.
→ Compare AAPL vs MSFT: see their actual pairwise correlation in AIQ
Historical Correlations Across Asset Classes
Long-run historical correlations (US-centric, 20-year periods) provide a baseline for diversification analysis. US large-cap equities and US investment-grade bonds: -0.2 to +0.1 (the correlation was positive in the 1970s-1990s inflation era, turned negative in the 2000s-2020s, and became positive again in 2022's simultaneous rate-inflation shock). US and international developed equities: 0.7-0.8. US and emerging market equities: 0.6-0.7. US equities and gold: 0.0 to -0.1. US equities and REITs: 0.55-0.70.
These long-run averages are useful baselines but can be deeply misleading if applied as static estimates. The equity-bond correlation, for example, flipped from positive to negative and back to positive over this period — the diversification benefit of the 60/40 portfolio was completely different in 2015 (strong bond diversification) than in 2022 (bonds and equities fell simultaneously). The correlation regime depends critically on the dominant macro driver: when inflation is the primary risk, equities and bonds correlate positively. When growth is the primary risk, they tend to correlate negatively.
Crisis Correlation: Why Diversification Fails When Needed Most
The most important — and most dangerous — property of correlation is that it is not stable during market crises. In normal, calm market conditions, correlations between sectors within US equities run 0.4-0.6, providing meaningful within-equity diversification. In acute crises (March 2020, October 2008), those correlations spike toward 0.85-0.95 as panic selling overrides all sector-specific dynamics. The portfolio that appeared well-diversified in the preceding months suddenly behaves like a single concentrated position.
This correlation contagion has a clear mechanism: during crises, forced selling (margin calls, redemption pressures, risk model breaches) affects all liquid equities simultaneously. Algorithmic risk reduction programs sell across all positions proportionally. International equity markets, which have 0.65-0.75 normal-period correlation with US equities, see that correlation rise to 0.90+ during global crises because the common driver — global risk aversion and dollar liquidity — dominates all local factors. The mathematical implication: equity diversification (whether by sector, geography, or market cap) provides less protection precisely when most needed.
True crisis diversification requires assets that maintain negative or near-zero correlation specifically during equity bear markets — not just during calm periods. US Treasuries, gold, and long volatility strategies have historically maintained or increased their negative equity correlation during crises. These are genuine portfolio crisis hedges; sector and geographic equity diversification is not.
→ Open Portfolio Attribution to find your portfolio's hidden risk clusters
Practical Application Checklist
- 1. Compute pairwise correlations across holdings using rolling 52-week windows — a 3-year static estimate misses regime shifts.
- 2. Flag any pair with correlation above 0.80 as potentially redundant — evaluate whether both positions are necessary.
- 3. Group holdings into correlation clusters; apply concentration limits to the cluster's total weight, not each position individually.
- 4. Check whether bonds and other "diversifiers" are genuinely negatively correlated in your portfolio's equity-stress scenarios, not just in calm market periods.
- 5. Re-assess correlations when the macro regime changes (inflation vs. growth, risk-on vs. risk-off) — correlations are regime-dependent, not fundamental constants.
- 6. Use stress testing: model what happens to your portfolio if all equity correlations rise to 0.9 simultaneously — this is the crisis scenario your normal-period correlation estimates ignore.
Common Pitfalls
- Assuming sector labels imply low correlation: technology and healthcare may share growth-factor exposure that makes them 0.7 correlated despite being "different sectors."
- Using a single static correlation estimate: correlations change with market regimes — a calm-period estimate dramatically underestimates crisis-period correlation.
- Confusing low correlation with low risk: an asset can be uncorrelated with equities and still be very volatile (crypto, for example). Correlation reduces the portfolio risk contribution; it doesn't eliminate the standalone risk of the asset.
- Treating international equity as genuine crisis diversification: in global risk-off events, international equities have historically converged with US equities — geographic diversification does not substitute for cross-asset diversification.
Apply Correlation In AlgoVestIQ
Correlation FAQs
What is a good correlation between portfolio holdings?
Lower is better for diversification. Correlations between 0.0 and 0.5 provide meaningful diversification benefit — portfolio volatility is substantially reduced below the average of individual asset volatilities. Correlations above 0.8 mean assets are nearly redundant from a risk perspective — adding a second highly correlated position adds minimal diversification while increasing complexity. Negative correlations (stocks and bonds in risk-off periods) provide the most powerful diversification.
Why do all stocks seem to fall together during market crashes?
During acute market stress, correlations across equities spike toward 1.0. The mechanism: panic selling is indiscriminate, algorithmic risk reduction affects all positions simultaneously, and the common macro driver (fear, liquidity, forced deleveraging) overwhelms all sector-specific and company-specific dynamics. This correlation contagion means equity-only diversification provides little protection precisely when it is most needed.
What assets have negative correlation to stocks?
US Treasuries (long duration) historically maintain negative to near-zero correlation with equities during risk-off events due to flight-to-safety flows. Gold tends toward zero to mildly negative equity correlation, with particularly low correlation during inflationary crises. Long volatility strategies (long VIX options, tail-risk hedges) have strongly negative equity correlation. These are the assets that provide genuine crisis diversification rather than the merely-low correlations of international equities that still spike during global downturns.
How do I measure correlation between my holdings?
Calculate the Pearson or Spearman rank correlation between the daily (or weekly) return series of each pair of holdings over a rolling 52-week or 3-year window. Most portfolio analytics tools (including AIQ's portfolio features) compute pairwise correlations automatically. The key is using rolling windows rather than a single static estimate — correlation changes significantly across different market regimes, and the static estimate is often most misleading at exactly the moment it matters most.
Can two stocks in different industries still be highly correlated?
Yes — sector labels are often misleading guides to actual correlation. Stocks with similar factor exposures (both high-growth, high-beta momentum names) can be 0.7-0.85 correlated even across different sectors. Conversely, companies with offsetting business models (an airline and a fuel distributor) can have low correlation within the same broad 'industrials' label. Real correlation analysis requires computing actual return correlations, not inferring them from sector classification.
Learn the concept, then apply it with live AIQ signals, rankings, screeners, and stock comparison workflows.