Correlation between asset classes: how it works and why it shifts

Correlation is the number that says whether two investments tend to move together. It runs from +1, perfect lockstep, through 0, no consistent relationship, to −1, exact opposites. Every argument for owning more than one asset class rests on it: combining assets reduces risk only to the extent that they do not rise and fall at the same time.

The general case for diversification, including the two-asset volatility formula, is in Diversification: why spreading your investments reduces risk. This page is about the number itself and about asset classes in particular: how a correlation is calculated, what it does and does not tell you, what 12 years of monthly returns show across eight asset classes, and why the most important correlation in finance, the one between stocks and high-quality bonds, changed sign in 2022 and has stayed positive since.

What correlation measures

The statistic used almost everywhere is the Pearson correlation coefficient. For two series of returns it asks one question, period by period: when one asset did better than its own average, did the other also do better than its own average?

• If above-average months for one line up with above-average months for the other, the correlation is positive.
• If above-average months for one line up with below-average months for the other, it is negative.
• If there is no pattern, it is near zero.

Because each asset is compared with its own average, correlation is blind to whether returns were positive or negative overall, and blind to their size. A bond fund that moves 0.5% whenever stocks move 5% can have a correlation of +1 with them.

The top line is the sum of co-movements; the bottom line scales it by each asset's own variability, which is what pins the result between −1 and +1.

Calculating one by hand

A real example that surprises people

From January to June 2022, the long Treasury fund TLT fell in all six months and SPY fell in four of them. It feels as though they "moved together." Their monthly price returns were:

Run the same five steps on these six months and the correlation comes out at only about 0.16. Both averages were deeply negative (−3.7% for SPY, −4.1% for TLT), but around those averages the two did not keep step: March was SPY's best month and TLT's second-worst, and June was SPY's worst month and TLT's best. Correlation captured the month-to-month pattern and missed the fact that both lost heavily. That is precisely its limitation. Two assets with low correlation can still lose money in the same year.

The asset-class correlation matrix

The matrix applies the formula to monthly price returns for seven funds and Bitcoin over January 2015 to October 2026, 142 months. Each cell is the correlation of the row asset with the column asset; the diagonal is 1 because everything moves perfectly with itself. Reading it by neighborhood:

Stocks with stocks. SPY and EFA, US and developed international stocks, were at 0.83. Global stock markets share the same business cycle and the same investors. Owning both diversifies across countries and currencies, but not across asset classes.

Real estate is mostly equity. VNQ, a REIT fund, was at 0.74 with SPY and 0.69 with EFA. REITs own property, but they are listed companies, valued by the same stock market. Their 0.58 with AGG also shows sensitivity to interest rates.

Bonds with bonds. AGG and TLT were at 0.90. Both are dominated by interest-rate moves; TLT simply has more of them (see Bond duration and convexity: measuring interest-rate risk).

Stocks with bonds. SPY–TLT was 0.12 and SPY–AGG 0.36. These are the numbers that justify the classic stock-and-bond portfolio, and they hide a regime change, discussed below.

Gold stands apart. GLD's correlation with SPY was 0.08, the lowest stock pairing in the matrix, and around 0.4 with the two bond funds, consistent with gold responding to interest rates and the dollar more than to profits.

Commodities against bonds. DBC, a commodity futures fund, had −0.32 with TLT and −0.15 with AGG, the only negative cells. Commodity prices and inflation tend to rise together, and rising inflation pushes bond yields up and bond prices down.

Bitcoin. BTC had 0.35 with SPY and near-zero correlations with bonds and gold. Its monthly swings are so large (an annualized volatility of about 71% on monthly returns since 2014) that even a modest correlation with stocks carries a lot of shared risk.

Stocks and bonds: a correlation that flipped

The chart recalculates the SPY–TLT correlation each month over the previous 36 months. Over the whole span from 2005 to 2021 it sat below zero, touching a positive value only briefly in 2009 (at most +0.05). Its low point was −0.78 in the window ending January 2013. Then it climbed through 2022, crossed zero in August 2022, reached +0.67 at the end of 2024 and was +0.44 in October 2026.

Splitting the matrix period at the turn of 2022 makes the same point:

The full-period 0.12 is an average of two different worlds.

Why the sign can change

A widely used explanation looks at what kind of news is driving markets:

• Growth shocks move stocks and bonds in opposite directions. When investors fear a recession, they expect lower profits (stocks fall) and lower interest rates (bond prices rise). In 2008 SPY's price fell 38.3% while TLT's rose 28.3%. Through most of 2000–2021, inflation was low and stable, and growth news dominated.
• Inflation and interest-rate shocks move them in the same direction. When inflation rises and the central bank raises rates, bond prices fall as yields rise, and stocks fall because their future profits are discounted at a higher rate. CPI inflation reached about 9% in mid-2022, and the Fed raised the upper end of its target range 11 times between March 2022 and July 2023, from 0.25% to 5.50%. In 2022 SPY's price fell 19.5% and TLT's 32.8%.

Under this explanation, the correlation stays positive as long as inflation and rate expectations drive both markets. The data since 2022 is consistent with that: consumer inflation was 3.7% year over year in September 2026, and the Fed's target range moved both down and, in September 2026, up again. The explanation is a framework, not a law, and no one knows in advance when a regime will change.

Correlation in a crash

Correlations measured in calm times also tend to understate how much risky assets move together in a panic. In October 2008, the worst month of the 2007–2026 period for most of the funds here, SPY fell 16.5%, EFA 20.8%, VNQ 31.7%, DBC 25.1% and even gold's GLD 16.1%, as investors sold what they could to raise cash. Treasuries were the exception that month: TLT fell 2.2%, and AGG 2.7%. In March 2020, SPY fell 13.0% while TLT rose 6.2%, a growth shock behaving as the framework predicts.

Pitfalls in using correlations

Sample length. Six months produced 0.16 above; 142 months produced 0.12; rolling 36- month windows ranged from −0.78 to +0.67. Short samples are noisy, and long samples can average away the regime that matters now.

Return frequency. Daily, weekly and monthly correlations for the same pair differ. This page uses monthly returns; Diversification: why spreading your investments reduces risk uses weekly ones and gets slightly different numbers for similar pairs.

Correlation is not magnitude. A correlation of 0.36 between SPY and AGG says nothing about the fact that AGG moves far less. For how much one asset moves per unit of another, use beta (Beta) or look at volatilities.

Correlation is not causation. Two assets can correlate because a third force (interest rates, the dollar, a recession) drives both.

Price-only data. These figures use price returns. Adding interest and dividends barely changes monthly correlations, but it changes whether a pair made money, which correlation never tells you anyway.

Correlation compared with related measures

R-squared, for a simple two-series case, is the correlation squared: SPY and EFA's 0.83 means about 69% of EFA's monthly variance lined up with SPY's.

How Finabase does it

Finabase does not calculate correlations or show a correlation matrix. What it does show are the prices that go into one: every fund in the matrix has a symbol page with a chart, for example SPY, TLT, AGG, GLD and VNQ, and Bitcoin is at BTC.X. The market lists for Treasury yields, futures and crypto put 1-week to 2-year returns side by side, which is a quick way to see whether a group of assets has been moving in the same direction.

The matrix and rolling chart on this page are ours, calculated from Finabase's frozen monthly price data for these funds, and the economic figures come from public Federal Reserve and BLS series.

Frequently asked questions

What is a good correlation for diversification?

The lower the better for reducing risk, with anything below about 0.3 offering meaningful diversification and negative values the most. But a low historical number is not a promise; SPY and TLT went from −0.27 to +0.56 between two periods.

Do stocks and bonds always move in opposite directions?

No. Over 2005–2021 the rolling correlation between SPY and long Treasuries was almost always negative, but it turned positive in August 2022 and was +0.44 in October 2026. In 2022 both fell, by 19.5% and 32.8% in price.

Which asset class has the lowest correlation with stocks?

In the January 2015 to October 2026 monthly data, gold (GLD) had the lowest correlation with SPY at 0.08, followed by long Treasuries (TLT) at 0.12. Cash, which barely moves, also diversifies stocks, though its correlation is less meaningful because its returns hardly vary.

Why do correlations rise in a crisis?

In a panic, investors sell whatever they can to raise cash or reduce risk, and that selling hits many assets at once. In October 2008, stocks, REITs, commodities and gold all had their worst month of the 2007–2026 period.

Summary

Correlation measures whether two assets tend to be above or below their own averages at the same time; it ignores the size of moves and whether either asset made money. Across asset classes from 2015 to 2026, stocks correlated highly with other stocks and with REITs, bonds with bonds, while gold, long Treasuries and commodities offered the most independence from stocks. The stock-bond correlation, negative through most of 2005–2021, turned positive in 2022 as inflation and rate increases hit both markets, a reminder that correlations are features of an economic regime and must be checked rather than assumed.