Key takeaways
- Over the 126 years from 1900 to 2025, US equities returned 6.6% a year after inflation, long bonds 1.6% and Treasury bills 0.5%, on the Dimson, Marsh and Staunton data UBS publishes.
- Across 16 countries from 1870 to 2015, the compounded real return was 6.61% a year for housing, 4.64% for equities, 1.94% for bonds and 0.78% for bills.
- Gold's real return was 1.3% a year across the whole of 1900 to 2025, but 4.7% a year for a US investor in the 54 years after Bretton Woods ended.
- From 1928 to 2025, $100 in the S&P 500 became $1,157,598.95 while US consumer prices rose 18.7 times, leaving a compounded real return of 6.78% a year.
- Housing's 6.61% is the most contested row: four Oxbridge college portfolios from 1901 to 1983 imply net real residential returns of about 2.3% a year.
Equities returned 6.6% a year after inflation. Cash returned 0.5%.
What has each asset class actually returned over the longest run anyone has measured, after inflation, from a source you can open yourself? Long-run asset class returns get quoted constantly and sourced rarely, so every row on this page names the study behind it.
Here it is for the United States. From 1900 to 2025, equities returned 6.6% a year in real terms, long bonds 1.6% and Treasury bills 0.5%. Those are the annualised rates in the Dimson, Marsh and Staunton database, which UBS publishes each year in the Global Investment Returns Yearbook and which now covers 35 markets. In cash terms, the same 126 years turned USD 1 into USD 124,854 in equities, USD 284 in long bonds and USD 69 in bills. The inflation index ended at USD 38. Divide one by the other and the equity dollar bought about 3,286 times as much as it did in 1900.
The distance between equities and bills is the equity risk premium, the extra return you were paid for holding the risky thing instead of cash. Over those 126 years it came to 6.1 percentage points a year. It is the most consequential number in long-run finance and it also carries the widest error bars, for reasons the last three sections deal with.
The table: long-run asset class returns, by study and by period
Every figure below is a real return after inflation, not a headline nominal number, and the historical returns by asset class come from four separate studies. No single dataset covers everything. Equities, bonds and bills go back furthest. Housing was largely missing from the record until 2017. Gold's record splits in two at the end of the Bretton Woods system, and the Yearbook reports the halves separately for that reason. So every row below names its study and its period, and keeps compounded returns separate from simple annual averages.
That last distinction does real work. In the 16-country data, equities averaged 6.89% a year in real terms and housing 7.05%, which makes them look like near neighbours. Compound the same series and equities fall to 4.64% while housing only falls to 6.61%. Nothing changed except volatility. The standard deviation of the annual equity numbers is 21.94% against 9.98% for housing, and volatility eats compounded returns. The arithmetic mean flatters a bumpy asset by 2.25 percentage points here, and it is the number you cannot actually earn.
| Asset class | Real return a year | Basis | Period | Source |
|---|---|---|---|---|
| US equities | 6.6% | compounded | 1900 to 2025 | DMS, in the UBS Yearbook |
| US long bonds | 1.6% | compounded | 1900 to 2025 | DMS, in the UBS Yearbook |
| US Treasury bills | 0.5% | compounded | 1900 to 2025 | DMS, in the UBS Yearbook |
| Gold, in US dollars | 1.3% | compounded | 1900 to 2025 | DMS, in the UBS Yearbook |
| Housing, 16 countries | 6.61% | compounded | 1870 to 2015 | Rate of Return on Everything |
| Equities, 16 countries | 4.64% | compounded | 1870 to 2015 | Rate of Return on Everything |
| Bonds, 16 countries | 1.94% | compounded | 1870 to 2015 | Rate of Return on Everything |
| Bills, 16 countries | 0.78% | compounded | 1870 to 2015 | Rate of Return on Everything |
| Housing, 16 countries | 7.05% | annual average | 1870 to 2015 | Rate of Return on Everything |
| Equities, 16 countries | 6.89% | annual average | 1870 to 2015 | Rate of Return on Everything |
| UK equities | 7.20% | annual average | 1871 to 2015 | Rate of Return on Everything |
| UK housing | 5.36% | annual average | 1900 to 2015 | Rate of Return on Everything |
| UK bonds | 2.29% | annual average | 1870 to 2015 | Rate of Return on Everything |
| UK bills | 1.16% | annual average | 1870 to 2015 | Rate of Return on Everything |
| US small caps | 8.67% | compounded | 1928 to 2025 | Damodaran and BLS, derived |
| S&P 500 | 6.78% | compounded | 1928 to 2025 | Damodaran and BLS, derived |
| Baa corporate bonds | 3.49% | compounded | 1928 to 2025 | Damodaran and BLS, derived |
| Gold | 2.50% | compounded | 1928 to 2025 | Damodaran and BLS, derived |
| US 10-year Treasuries | 1.46% | compounded | 1928 to 2025 | Damodaran and BLS, derived |
| US house prices, no rent | 1.13% | compounded | 1928 to 2025 | Damodaran and BLS, derived |
| US 3-month Treasury bills | 0.33% | compounded | 1928 to 2025 | Damodaran and BLS, derived |
One pattern survives every cut of the data. Riskier assets paid more, in the order you would expect, and the ordering held almost everywhere. Equities were the best-performing asset class in all 21 Yearbook countries with continuous investment histories, and bonds beat bills in every country except Portugal.
Gold's long run hides two completely different regimes
Gold returned 1.3% a year in real dollar terms across the full 126 years, turning one unit of purchasing power into 5.2. On its own, that number is close to useless.
The Yearbook doesn't leave it at one number either, and splits the record at the end of the Bretton Woods system. In the 54 years since, real gold returns ran at 4.7% a year for a US investor, 5.8% for a UK investor and 4.3% for a Swiss one. Same metal, same century, and a rate more than three times the full-period figure depending on which half of the record you stand in. The 1928 to 2025 window in the table sits between the two at 2.50%, because it contains both regimes.
The Yearbook also tests the standard claim that gold hedges inflation, and doesn't find much. Of the 28 years in which inflation exceeded 3%, gold's return was negative in 13. That's a coin flip dressed as a hedge. If the allocation question interests you more than the return series, the evidence on how much gold in a portfolio is a separate piece.
Housing looks like the best long-run asset in the data, and it's the most contested row
The 16-country database assembled by Jorda, Knoll, Kuvshinov, Schularick and Taylor, published as a San Francisco Fed working paper in December 2017, was the first to put housing on the same footing as stocks and bonds. It runs from 1870 to 2015 and it finds housing compounding at 6.61% a year in real terms against 4.64% for equities. The authors state the implication plainly: "residential real estate, not equity, has been the best long-run investment over the course of modern history."
The risk-adjusted version is starker still. Housing delivered a higher return per unit of risk in all 16 countries, with Sharpe ratios "on average more than double those of equities". The mechanism sits in the income. In nominal terms, compounded housing returns split almost evenly between 5.25% from prices and 5.47% from rent. Equities split 4.46% from prices and 4.16% from dividends. Rent is the steadier half of a steadier asset, and steadiness is what compounds.
The counter-case is serious. David Chambers, Christophe Spaenjers and Eva Steiner hand-collected property-level accounts for the institutional portfolios of four large Oxbridge colleges covering 1901 to 1983. Gross income yields started around 5%, real income growth was close to zero, and their conclusion is that real estate is "a less profitable investment in the long run than previously thought". Their estimate of annualised real total returns net of costs runs from about 2.3% for residential property to 4.5% for agricultural. That's 4.31 percentage points below the 16-country housing figure, and the phrase carrying the weight is "net of costs".
The 16-country study makes its own cost estimate and lands in between. Round-trip transaction costs across 13 of its 16 countries average about 7.7% of a property's value, but people hold houses for a long time, so the authors put the drag at 77 basis points a year. On the same arithmetic, equity transaction costs came to at least 100 basis points a year over a century, because turnover is so much higher. Take 77 basis points off housing and it still compounds at 5.84%. The two studies are not measuring the same thing: one is a national index, the other is four specific portfolios with real ledgers behind them.
Neither number describes what a household owns. Both track an index or a diversified institutional portfolio, and one house is neither. It has a postcode, a roof and a tenant, none of which diversify. The 16-country figures are also unleveraged, while most households hold property on a mortgage. The country-by-country comparison of housing versus equities goes further into that.
US property returned 6.03% and 1.13% a year in two datasets, and both are right
Here's the discrepancy that catches people out, and it's the most useful thing on this page.
In the 16-country database, US housing returned 6.03% a year in real terms as an annual average from 1891 to 2015. In Aswath Damodaran's US dataset at NYU Stern, running from 1928 to 2025, the same asset class compounds at 1.13% a year in real terms. Same country, overlapping century, roughly five percentage points apart.
Damodaran explains it in the workbook itself. His real estate row uses Robert Shiller's home price series, which became the Case-Shiller index, and he writes that "this return is just price appreciation, and will understate the returns on real estate by the cash flow return (from rental income) each year". The 16-country series is a total return, price plus rent, and rent supplied 5.47 of its 10.73 percentage points of nominal return. Strip the rent out and most of the gap closes. The rest is the arithmetic-versus-compounded difference again, since 6.03% is an average of annual returns and 1.13% is a compounded rate.
The lesson generalises well past housing. Every published figure for long-run asset class returns is really four statements: which index, total return or price only, arithmetic or compounded, and deflated by which price series. Change one and the answer moves by more than the gap between two different asset classes.
Ninety-eight years of US data, computed from one dataset and one price index
The chart above is the cleanest single comparison on this page, because every bar comes from one return series and one deflator. Damodaran's dataset records what $100 invested at the start of 1928 was worth at the end of 2025 if it was left alone. The S&P 500 with dividends reinvested reached $1,157,598.95. Small caps reached $6,462,598.52, Baa corporate bonds $53,952.41, gold $21,025.41, ten-year Treasuries $7,752.88, house prices $5,626.02 and three-month Treasury bills $2,578.30.
US consumer prices over the same span went from a December 1927 CPI-U reading of 17.3 to 324.054 in December 2025, a rise of 18.7 times, or 3.04% a year. Deflating each column by that gives the compounded real returns in the chart: 8.67% for small caps, 6.78% for the S&P 500, 3.49% for Baa corporate bonds, 2.50% for gold, 1.46% for ten-year Treasuries, 1.13% for house prices and 0.33% for Treasury bills.
Two things are worth pausing on. The S&P 500 figure lands within 0.2 percentage points of the 6.6% the DMS data gives for 1900 to 2025, over a different period and a different index. Independent agreement that close is rare in this field, and it's the main reason to trust the top of the table. Second, bills earned 0.33% a year for 98 years. Cash didn't lose to inflation over the long run. It just barely beat it, and it did so before any tax.
The averages hide a spread wide enough to matter more than the averages
Most countries cluster. Long-run risky asset returns in the 16-country set are "close to 6%-8% per year", which is a tight band for a century and a half. The tails are not tight at all. Real equity returns over the full sample ran from 3.25% a year in France to 9.98% in Finland. That is not a rounding difference. Compounded over 40 years, the French rate multiplies real wealth 3.59 times and the Finnish rate 44.93 times.
Then there is the path. The Yearbook notes that since 1900 equities and bonds have each, on several occasions, lost more than 70% in real terms, while a 60:40 blend of the two has never fallen more than 50%. An annualised return is a summary of a path, and the path is the part a person has to live through. How long those recoveries took is its own question, and the honest answer on bear market recovery time is longer than most people assume.
Diversification is also doing less work than it used to. By the end of 2025, US equity market concentration was at its highest level for at least 100 years, and the US alone accounted for around 62% of world equity market value. Correlations between developed and emerging markets, and between equities and bonds, have risen with it. What that does to a portfolio when it matters is covered in the evidence on diversification in a crisis.
And the developed-versus-emerging ranking flips depending on the start date. Developed markets returned 8.5% a year against 6.9% for emerging markets across the whole period. From 1960 onwards, emerging markets won.
What this evidence cannot tell you
Every figure on this page is a backtest. None of them is a forecast, and the arithmetic of a compounded average says nothing about the order the years arrive in.
There is a selection question sitting underneath the headline claim. The Yearbook's strongest statement, that equities beat everything, is scoped to the 21 countries "with continuous investment histories". That qualifier is doing work, because a market that stopped having a history is a market that took an unrecoverable loss. The 16-country sample is all advanced economies, chosen partly because their records survive. Both datasets are honest about their coverage, and both are still samples of the winners' shelf.
The periods do not line up. The three blocks in the table start in 1900, 1870 and 1928 and end in 2025, 2015 and 2025. The 16-country housing series for the UK only begins in 1900, and its US series in 1891. Comparing a row from one block against a row from another is a caveat, not a calculation.
Currency is missing from the US rows. They are dollar returns deflated by US inflation, which is not what a sterling investor experienced. The Yearbook's own estimate is that currency risk added around 6 percentage points to total risk, for equity and bond holders alike.
Tax is missing from all of them. The 16-country authors say so directly: "When calculating equity and housing returns, we do not account for taxes." The housing rows are net of maintenance, ground rent and other irrecoverable costs inside the rental yield, and gross of everything else.
What would change the conclusion
If the equity risk premium turns out to be a feature of one sample rather than a fact about risk, the top of the table goes with it. The 6.1 percentage point gap between US equities and US bills rests on one path through one century in one country that spent that century winning. The 16-country compounded figure of 4.64% is the same claim, measured more broadly, and it is nearly two points lower.
If the housing income data are revised, the ranking changes. The 16-country housing series is built from a rent-price ratio benchmarked in one year and extended backwards. The Oxbridge college numbers are ledgers. More archival work landing nearer 2.3% than 6.61% would move the top row of this table below equities, and the "housing beats stocks" finding would become a footnote about method.
If inflation behaves differently, every real number here moves. Each one is a nominal series minus a price index, and the 1928 to 2025 US figures rest on a single CPI-U series compounding at 3.04% a year. A different deflator gives different real returns from identical nominal data.
The thing worth watching isn't next year's return. It's whether the next revision to one of these datasets moves a row by more than the distance between two rows. Long-run asset class returns are a measurement, not a constant, and on housing that measurement has already moved once.