Measure and manage portfolio risk Lesson 5 of 14
Portfolio and risk · Measuring return and risk
Correlation in a portfolio: what a correlation matrix tells you
How correlation between holdings sets a portfolio’s volatility: the two-asset formula, a matrix of eight funds since 2005, and what changed in 2008 and 2022.
On this page
Correlation measures how consistently two investments move in the same direction, on a scale from −1 to +1. Inside a portfolio it is the number that decides how much risk is removed by holding several things instead of one. Two holdings that rise and fall together add their risks. Two that move independently cancel part of each other's.
The beginner lesson on diversification introduces the idea, and Correlation between asset classes: how it works and why it shifts surveys the numbers across asset classes. This lesson is about using correlation as a portfolio tool: the formula that turns weights, volatilities and a correlation into a portfolio's volatility, how to read a matrix of many holdings, and the uncomfortable fact that correlations are least reliable when they are needed most.
The idea
Put two holdings side by side and ask, month after month: when one did better than its own average, did the other? If the answer is nearly always yes, the correlation is near +1. If it is yes about half the time, the correlation is near zero. If it is nearly always no, the correlation is near −1.
For a portfolio, the consequence is direct. Bad months hurt most when everything has a bad month together. If two holdings take turns, each one's bad months are partly covered by the other's ordinary or good ones, and the combined account swings less than either would suggest. Correlation is the measure of how much taking turns to expect.
A correlation matrix shows this for every pair at once.
Read it like a distance chart between cities. Each cell is one pair; the diagonal is each fund with itself, always 1. The matrix is symmetric, so the upper-right half repeats the lower-left. Two regions stand out. The four stock funds and the real-estate fund form a block of high values, 0.60 to 0.91. The rows for Treasuries and gold are pale: their correlations with the stock funds are close to zero.
How it works
The first two terms are each holding's own risk, scaled by its weight. The third is the interaction, and it is the only place correlation appears. When ρ is +1 the formula collapses to the plain weighted average of the two volatilities: no risk is removed. Any ρ below +1 makes the result smaller than that average. A negative ρ makes the third term negative, subtracting risk.
A worked example: 60% SPY, 40% AGG
The inputs, from monthly total returns over January 2005 to October 2026: SPY volatility 14.77%, AGG volatility 4.49%, correlation 0.23.
- SPY's own term: 0.6² × 14.77² = 0.36 × 218.2 = 78.5.
- AGG's own term: 0.4² × 4.49² = 0.16 × 20.2 = 3.2.
- Interaction: 2 × 0.6 × 0.4 × 14.77 × 4.49 × 0.23 = 7.3.
- Variance: 78.5 + 3.2 + 7.3 = 89.1. Volatility: the square root, 9.44%.
The actual volatility of that portfolio, rebalanced every month, was 9.44%. The formula is exact, not an approximation. The weighted average of the two volatilities is 0.6 × 14.77 + 0.4 × 4.49 = 10.66%, so correlation below 1 removed about 1.2 points.
Changing only the correlation shows how much it matters here:
| Correlation between SPY and AGG | Volatility of the 60/40 mix |
|---|---|
| +1.0 | 10.66% |
| +0.6 | 10.04% |
| +0.23 (actual) | 9.44% |
| 0 | 9.04% |
| −0.3 | 8.50% |
The range is narrower than might be expected: from fully correlated to mildly negative, the portfolio's volatility moves by about two points. The reason is in step 1. Stocks account for 78.5 of the 89.1 units of variance, or 88% of the portfolio's risk, with 60% of the money. Bonds are too calm for their correlation to change much. A 60/40 portfolio is mostly a smaller stock portfolio. Most of its risk reduction comes from holding less stock, not from the correlation. Risk parity: weighting a portfolio by risk instead of by money starts from that observation.
Correlation does far more work when the two holdings have similar volatility. Half SPY and half GLD, with volatilities of 14.8% and 17.4% and a correlation of 0.09, has a weighted average volatility of 16.1% and a formula volatility of 11.9%, lower than either holding on its own.
More than two holdings
With many holdings the same formula has one own-risk term per holding and one interaction term per pair. Ten holdings have 10 own terms and 45 pairs; fifty have 50 and 1,225. As a portfolio grows, the pairs dominate, and its risk is determined almost entirely by the average correlation between holdings, not by any one holding's volatility. The mathematics of diversification: how many holdings are enough takes that to its limit.
That is why the average of the matrix matters. The five stock-like funds (SPY, QQQ, EFA, IWM, VNQ) have an average pairwise correlation of 0.77. Owning all five feels diversified and behaves much like owning one. Across all eight funds the average is 0.39, pulled down entirely by the three that are not stocks.
Using it: what the long-run number hides
Correlations change with the period
Measured over the previous 36 months, the correlation between SPY and AGG was −0.47 at the end of 2012 and +0.73 at the end of 2024. The full-period figure of 0.23 is the average of two regimes that had little in common. From 2008 to 2020, falling inflation and falling interest rates meant bonds tended to rise when stocks fell. From 2022, rising inflation and rising rates hurt both at once. Gold's three-year correlation with SPY moved between −0.27 and +0.44 around an average near zero.
Two calendar years
| 2008 | 2022 | ||
|---|---|---|---|
| SPY | −36.8% | SPY | −18.2% |
| AGG | +7.9% | AGG | −13.0% |
| TLT | +34.4% | TLT | −30.8% |
| GLD | +4.9% | GLD | −0.8% |
| EFA | −41.1% | EFA | −14.2% |
| VNQ | −37.1% | VNQ | −26.2% |
In 2008 a holder of long Treasuries gained 34.4% while stocks lost 36.8%. In 2022 the same fund lost 30.8%, more than stocks. An investor who had read "Treasuries are negatively correlated with stocks" as a permanent fact was protected in the first year and hurt in the second. Both years are in the data behind the full-period correlation of −0.09 between SPY and TLT.
Correlations rise when markets fall
Splitting the 262 months by whether SPY rose or fell:
| Correlation with SPY | In the 89 months SPY fell | In the 173 months SPY rose |
|---|---|---|
| EFA (foreign stocks) | 0.82 | 0.64 |
| IWM (small companies) | 0.83 | 0.73 |
| AGG (bonds) | 0.22 | 0.14 |
| GLD (gold) | 0.04 | 0.08 |
Stock funds were more tightly linked in falling months than in rising ones, a pattern documented for international markets by Longin and Solnik in 2001. In the 27 worst months for SPY, those with a loss of 4.7% or more, foreign stocks fell in 26 and small companies in all 27, by an average of 8.1% and 9.5% against SPY's 7.5%. Spreading money across kinds of stocks removed almost nothing in those months. Over the same 27 months AGG averaged −0.3%, long Treasuries +1.3% and gold +1.1%; they were down in 16, 13 and 11 of the 27 months respectively.
The practical reading: diversification among stocks protects against one company or one sector failing, not against a market-wide fall. Protection in a fall has come from assets that are not stocks, and even that has depended on the kind of crisis.
Where it fails or misleads
- Correlation is not size. A holding can be 0.9 correlated with the market and move a third as much, or 0.3 correlated and move three times as much. The combination of correlation and relative volatility is beta.
- It measures straight-line relationships. Two assets can be unrelated in ordinary months and fall together in a crisis. The average correlation will be low, and the portfolio will not be protected when it matters.
- Zero is not negative. An asset with zero correlation does not offset losses. It simply does its own thing. Only a negative correlation, or holding less of the risky asset, cushions a fall.
- Short windows are noisy. With 36 monthly observations, a measured correlation of 0.2 is statistically indistinguishable from zero. Differences of 0.1 or 0.2 between pairs should not drive decisions.
- Sampling interval matters. Daily correlations between assets that trade in different time zones understate how related they are; monthly returns are a safer basis.
- A low correlation does not make a bad asset good. Adding something with a low correlation and a negative expected return lowers risk and lowers return. Correlation describes how risks combine, not whether a holding is worth owning.
Compared with its neighbours
| Measure | Range | Answers | Used for |
|---|---|---|---|
| Correlation | −1 to +1 | Do they move in the same direction, consistently? | How risks combine in a portfolio |
| Covariance | Any number | The same, in units of return squared | The raw input to portfolio formulas |
| Beta | Any number | How much does one move per 1% move of the other? | Exposure to a market or a hedge ratio |
| R-squared | 0 to 1 | What share of one's variation does the other explain? | How much a benchmark explains a fund |
How Finabase does it
Finabase does not calculate a correlation between two symbols or show a correlation matrix for a portfolio.
- The comparison chart is the closest tool. It draws up to several symbols as percent changes from a common starting bar. Lines that rise and fall in step are highly correlated; lines that cross repeatedly are not. It gives a picture, not a number.
- The chart's CORREL study is a different calculation that uses a single symbol's own price series; Correlation coefficient (CORREL): how two series move together explains what it measures. It does not compare two symbols.
- An ETF's Holdings tab lists its largest positions, the weight of the top ten and its sector and country weights. Two funds with heavily overlapping holdings will be highly correlated whatever their names suggest.
- Portfolio pages show each position's weight in an allocation bar, with no sector breakdown and no correlation or risk figure.
For an actual matrix, export month-end prices of the holdings from each symbol's History
tab, compute the monthly percent changes, and use a spreadsheet's CORREL function for each
pair. Use at least five years of months.
Frequently asked questions
What correlation counts as diversifying?
There is no threshold, because any correlation below +1 reduces risk a little. As a rough guide, holdings correlated above 0.8 behave as one, and the benefit becomes noticeable below about 0.5. What matters as much is whether the correlation holds in bad markets.
Why do all my stock funds fall together?
Because they are highly correlated with each other. Five stock and real-estate funds averaged 0.77 over 2005 to 2026, and more than that in falling months. They share one large risk: the stock market itself.
Are stocks and bonds negatively correlated?
Sometimes. SPY and AGG's three-year correlation has been as low as −0.47 and as high as +0.73. Over the whole of 2005 to 2026 it was +0.23. The sign has depended on whether inflation or growth was the market's main worry.
Is a negative correlation always better?
It reduces risk more. But assets with reliably negative correlations to stocks tend to have low or negative expected returns, since everyone would like to own them. Portfolio insurance has a cost; see Hedging a portfolio: four ways to reduce risk and what each costs.
How many holdings does it take to diversify?
It depends on their correlations, not their number. Thirty stocks from one industry diversify less than four funds from different asset classes.
Summary
Correlation is the term in the portfolio-variance formula that decides whether risks add or partly cancel. For a 60/40 mix of SPY and AGG, a correlation of 0.23 gave a volatility of 9.44%, exactly what the two-asset formula predicts and 1.2 points less than the weighted average. The matrix of eight funds shows a tight block of stock-like holdings and three assets that stand apart. Those relationships are averages: the stock–bond correlation has swung from −0.47 to +0.73, stock funds move closer together in declines, and bonds helped in 2008 and hurt in 2022. Correlation is essential for understanding a portfolio and unreliable as a promise about its next bad year.
Check your understanding
5 questions on this lesson. Each answer comes with the reason, and nothing is scored.
Two funds each have volatility of 20%, and a portfolio holds half of each. If their correlation is +1, the portfolio’s volatility is 20%. What is it if the correlation is 0?
- 20%
- About 14%
- 0%
Show the answer
B. About 14% Variance = 0.25 × 400 + 0.25 × 400 + 0 = 200, and the square root is 14.1%. Zero correlation removes part of the risk, not all of it; only a correlation of −1 with equal weights and volatilities would bring it to zero.
A 60/40 stock and bond portfolio’s volatility changes only from 10.7% to 8.5% as the stock–bond correlation goes from +1 to −0.3. Why so little?
- Correlation does not affect portfolio risk
- Stocks are so much more volatile than bonds that they supply nearly all of the portfolio’s risk
- The formula only works for two stocks
Show the answer
B. Stocks are so much more volatile than bonds that they supply nearly all of the portfolio’s risk With stock volatility near 15% and bond volatility near 4.5%, the stock term is about 88% of the portfolio’s variance. The interaction term that carries the correlation is small, so changing the correlation changes little.
Fund X has a correlation of 0.9 with the market. What does that tell you about how far it falls when the market falls 10%?
- It falls about 9%
- Nothing about the size; correlation measures consistency of direction, not magnitude
- It falls about 10%
Show the answer
B. Nothing about the size; correlation measures consistency of direction, not magnitude A fund can be 0.9 correlated and move a third as much or twice as much. The size of the move is given by beta, which is correlation times relative volatility.
An investor holds five different stock funds whose average correlation with each other is 0.77. How diversified is the portfolio against a market-wide fall?
- Well diversified: five funds spread the risk five ways
- Barely: the funds share one dominant risk and have fallen together in the market’s worst months
- Fully: correlations below 1 remove market risk
Show the answer
B. Barely: the funds share one dominant risk and have fallen together in the market’s worst months High pairwise correlations mean the holdings behave as one. In SPY’s 27 worst months since 2005, small-company stocks fell in all 27 and foreign stocks in 26.
Long Treasuries gained 34% in 2008 and lost 31% in 2022, while stocks fell in both years. What does this show about correlation?
- The correlation was measured wrongly in one of the years
- A correlation is an average over a period and can change sign between regimes
- Treasuries are always a poor hedge
Show the answer
B. A correlation is an average over a period and can change sign between regimes The stock–bond relationship was negative while falling inflation and rates dominated, and positive when rising inflation and rates hurt both. A long-run figure near zero described neither year.
Exercise
One task to do in Finabase. Nothing is graded or saved.
Open the Sectors page and find “Sector index performance”. In the Today list, count how many of the 11 sectors have the same sign. Then read the 1 year list and note the gap between the best and the worst sector. Do sectors that move in the same direction on the day end up with similar returns over the year?
Open the Sectors pageWhat you should find
On most days a clear majority of sectors share one sign, and on a sharp market fall nearly all do. Stock sectors are highly correlated, so holding several of them is close to one bet when the market drops. The one-year gap between best and worst is still wide, because correlation measures direction and consistency, not the size of the moves.
Sources and further reading
- Harry Markowitz, “Portfolio Selection,” Journal of Finance, vol. 7, no. 1 (1952)
- François Longin and Bruno Solnik, “Extreme Correlation of International Equity Markets,” Journal of Finance, vol. 56, no. 2 (2001)
- Zvi Bodie, Alex Kane and Alan J. Marcus, Investments, 12th ed. (McGraw-Hill, 2021) — portfolio variance with two risky assets
- Finabase chart data — monthly bars and dividends of SPY, QQQ, EFA, IWM, AGG, TLT, GLD and VNQ, January 2005 to October 2026; correlations computed by Finabase
