Does correlation change with your time horizon? 98 years of US stocks, bonds and gold

Correlation measured on monthly or yearly returns does not tell you which assets kept their value over the decades that ruin a long-term plan. In 98 years of US data, stocks lost purchasing power over 11 of 89 rolling ten-year windows.

In 8 of those 11, 10-year Treasury bonds lost purchasing power too, mostly in the inflation of the 1960s and 1970s; gold gained in 10. For a household investing for decades, the useful check is how its assets behaved over long stretches, and under which kind of inflation, not only how they moved from one month to the next.

Key takeaways

1. Correlations change with the horizon. In real US returns from 1928 to 2025, the stock-gold correlation was -0.07 on one-year returns and -0.68 on overlapping ten-year returns; the stock-bond correlation went from 0.09 to 0.21.
2. Lost decades for stocks happened, and clustered. US stocks lost purchasing power over 11 of 89 ten-year windows, starting in 1937, from 1965 to 1973 and from 1999 to 2001.
3. Bonds protected in one of three stretches. Treasuries gained 4.0% a year after inflation from 1999 to 2008, but lost in 8 of the 11 losing windows, mostly when inflation was high.
4. Inflation decides the stock-bond relationship. The annual correlation of real stock and bond returns was 0.40 in years with inflation above 4% and -0.15 in the others.
5. The long-run cost of safety was large. Stocks compounded at 6.8% a year after inflation over the full period, a 60/40 stock-bond mix at 5.1% and bonds at 1.5%.

Why the horizon matters

Most published correlations use daily, weekly or monthly returns. They describe how assets move together over short periods, which matters for a sale next month. A household building wealth over 20 or 30 years also needs to know something different: whether its assets can lose purchasing power over the same long stretch.

Two assets can be uncorrelated month to month and still lose together over a decade, if both are hurt by the same slow force, such as rising inflation. The reverse is possible too.

The data

We used Aswath Damodaran's historical returns for the US, 1928 to 2025, published by NYU Stern: the S&P 500 with dividends, 3-month Treasury bills, 10-year Treasury bonds, Baa corporate bonds, US house prices and gold, all annual and in dollars. We converted every return to real terms with the US consumer price inflation in the same workbook, and calculated:

• correlations of real returns over 1, 5, 10 and 20-year windows;
• the real annualised return of every asset in each of the 89 rolling ten-year windows, and in a 60% stock, 40% bond mix rebalanced each year.

The data is American, in dollars and before costs and taxes. It is used here because it is one of the few public series long enough to contain several lost decades.

Correlation by horizon

Source: Copio calculation on Damodaran's data; correlations by horizon (CSV).

Three things change as the horizon lengthens:

• Gold's relationship with stocks becomes strongly negative. Year to year, gold and stocks were close to unrelated. Over ten years, gold tended to do well in the decades when stocks did badly, mainly the 1970s and 2000s.
• Bonds become slightly more aligned with stocks. Over long windows, both are hurt by sustained inflation, which pushes their real returns in the same direction.
• Corporate bonds become less aligned. Their year-to-year link with stocks comes from credit fear in recessions, which fades over a decade.

A warning on precision: overlapping ten-year windows share nine of their ten years, so 89 windows contain only about nine independent decades. The non-overlapping column shows how much a single estimate can move; house prices go from 0.17 to -0.41. Read long-horizon correlations as a description of a few historical episodes, not as stable parameters.

The decades when stocks lost purchasing power

US stocks, with dividends reinvested, lost purchasing power over 11 of the 89 ten-year windows. They fall into three stretches.

Source: Copio calculation on Damodaran's data; lost decades (CSV), all ten-year windows (CSV).

Each stretch had a different cause and a different winner:

• 1937 to 1946 combined a stock crash with wartime inflation and capped interest rates. Almost everything financial lost purchasing power, including bills and gold, whose dollar price was fixed by law. Corporate bonds and house prices edged ahead.
• 1965 to 1974, and the overlapping windows up to 1973 to 1982, covered the inflation era. Stocks and bonds lost together, and gold, freed from its fixed price in the early 1970s, gained 12.3% a year after inflation.
• 1999 to 2008 started with expensive stocks and ended in the financial crisis, with moderate inflation. Treasuries gained 4.0% a year after inflation and the 60/40 mix roughly held its purchasing power.

Across all 11 losing windows, the number in which each asset beat inflation was: gold 10, house prices 8, Baa corporate bonds 6, 10-year Treasuries 3, T-bills 3, and the 60/40 mix 3. Our 1970s study covers the inflation decade in more detail.

Inflation and the stock-bond relationship

The annual data shows the same split.

Source: Copio calculation; annual returns 1928-2025 (CSV).

In 26 of the 98 years, stocks lost money in nominal terms; bonds gained in 21 of them. Both fell in the same year only five times: 1931, 1941, 1969, 2018 and 2022. That is a good record for single bad years, and a weaker one for inflationary decades. Our monthly euro data shows the same break after 2021 in the falling-months study.

What diversification cost over 98 years

Over the full period, every alternative to stocks compounded more slowly, and the gap is large over a lifetime.

At 6.8% a year, purchasing power doubles in about 10.5 years; at 5.1%, in about 14. The case for holding other assets is not that they grow wealth faster, but that they can keep part of a portfolio intact through the decade when stocks do not, which matters most for someone who has to spend or sell during it. That trade-off is personal: it depends on when money is needed, on other income and on debts.

Checks you can run on your own plan

1. Name the decade your plan cannot survive. Write down what you would need to sell, and when, if your equities lost purchasing power for ten years.
2. Check each holding against the three stretches. Use the ten-year windows (CSV) to see how a similar asset fared in 1937 to 1946, 1965 to 1974 and 1999 to 2008.
3. Separate inflation risk from crash risk. Government bonds helped in deflationary or low-inflation crises and hurt in inflationary ones; ask which your portfolio is built for.
4. Weigh the cost. Compare the long-run real returns above with the protection each asset gave, rather than looking at either alone.
5. Use your own currency. These are US results in dollars. A euro investor faces different inflation, rates and exchange-rate effects.

Limits of this analysis

• The data is American, in dollars, before costs and taxes, and covers one country that turned out well over the century.
• Damodaran's bond returns are calculated from constant-maturity 10-year Treasury yields, his real estate series is house-price appreciation without rent, and his gold series uses year-end prices; US residents could not freely hold gold from 1933 to 1974, so its early history describes a price, not an investment most people could make.
• Ten-year windows overlap and rest on few independent decades; small changes in the window can move the results.
• Real returns use annual consumer price inflation, which may differ from a household's own inflation.
• History shows what happened in a few regimes. It does not forecast the next decade.

How can copio help you examine your own allocation?

Copio is in private beta across Europe, with early access by application. It looks through your funds into asset classes and places each on a return-correlation map as a main driver, a true diversifier, dead weight or a shock absorber, so you can see whether your holdings differ by driver or only by label.

A correlation matrix shows which holdings move together, and a crash test shows how a mix like yours would have fared in past crises. Copio provides educational analysis, not personalised advice; it does not place trades or choose an allocation for you. The free portfolio checker is a first step.

This article is published for information and educational purposes only. It does not constitute personalised investment advice. Past performance is not a reliable indicator of future performance.