Key takeaways
- Across the 21 countries with unbroken equity records from 1900 to 2014, annualised real returns ran from 0.6% a year in Austria to 7.4% a year in South Africa.
- The United States returned 6.5% a year after inflation over those 115 years. That was third of the 21, behind South Africa's 7.4% and Australia's 7.3%.
- Strip out dividends and start the clock in 1921 and the order changes. The US is then first of 39 markets at 4.3% a year, against a median of 0.75% across all 39.
- Of the 24 markets with price data in 1931, only seven traded without interruption. Ten suffered a long-term closure, and some never reopened.
- Dimson, Marsh and Staunton put the cost of omitting dead markets at roughly 0.1 percentage points a year on the world equity premium, not the several points often claimed.
Stock market returns by country ran from 0.6% to 7.4% a year
Ask what shares return over the long run and you'll usually be handed an American number. The honest answer is a range, and it's a wide one. Stock market returns by country have been measured back to 1900, and the gap between the best and worst national markets runs to nearly seven percentage points a year.
Over the 115 years from 1900 to 2014, the 2015 Credit Suisse Global Investment Returns Yearbook measured annualised real equity returns of 0.6% a year in Austria and 7.4% a year in South Africa. Both figures are total returns with dividends reinvested, after that country's own inflation. The same book puts the United States at 6.5%, the United Kingdom at 5.3%, and its 23-country world index at 5.2%.
Compounding turns that spread into something unrecognisable. A unit invested in Austrian shares in 1900 grew, in real terms, by a factor of 1.9 by the end of 2014. In South Africa the factor was 3,551. In the US it was 1,396 and in the UK 367. Same asset class, same 115 years, same reinvestment assumption.
So the first thing the cross-country record does is take the single number out of your hands. There isn't one long-run equity return. There's a distribution, and which end of it you landed on had rather more to do with the century your country had than with anything an investor chose.
The table: 21 countries, annualised real equity returns, 1900 to 2014
These are the countries the Yearbook carries with a continuous index from 1900. Russia and China are in the database too, but only as broken series, which a later section comes back to.
- Austria 0.6%, Italy 1.9%, Belgium 2.7%
- France 3.2%, Germany 3.2%, Portugal 3.4%, Spain 3.7%
- Japan 4.1%, Ireland 4.2%, Norway 4.2%, Switzerland 4.5%
- Netherlands 5.0%, Denmark 5.3%, Finland 5.3%, United Kingdom 5.3%
- Canada 5.8%, Sweden 5.8%, New Zealand 6.1%
- United States 6.5%, Australia 7.3%, South Africa 7.4%
The chart above plots the same 21 figures in order. The median country returned 4.5% a year, which is Switzerland. The US sat two percentage points above that median, and two countries sat above the US.
That last detail matters, because the usual framing of this data is that America is an outlier. On this measure it isn't quite. It's near the top of a long right tail, which is a weaker claim and a more defensible one.
The bottom of the table is a war record, not an investment record
The countries below 3.5% share a pattern, and it isn't about their companies. Austria's 0.6% is the lowest of any market with a 1900 record, and the Yearbook's own country note explains why: Austria-Hungary broke up after the First World War, the country suffered hyperinflation in 1921 and 1922, it was annexed in 1938, and its exchange reopened in 1948 to such thin trading that there wasn't a single flotation in the 1960s or 1970s.
Italy at 1.9%, Belgium at 2.7%, France and Germany at 3.2% tell versions of the same story. In Italy, Belgium and France the bondholders did worse than the shareholders, and in Austria they were wiped out. Austrian bonds returned -3.8% a year in real terms over the 115 years and Austrian Treasury bills -8.1%, because inflation did the work that default usually does.
Japan is the cleanest illustration that the average hides regimes. From 1900 to 1939 Japan was the world's second-best equity performer. Then, in the Yearbook's words, "World War II was disastrous and Japanese stocks lost 96% of their real value." From 1949 to 1959 Japanese equities gave a real return of 1,565%. By the start of the 1990s Japan was 41% of the world index against 30% for the US. Its weighting then fell from 41% to 8%. The 115-year average of all that is 4.1% a year, a number that describes none of it.
Change the dataset and the US goes from third to first
Philippe Jorion and William Goetzmann built the other famous cross-country database, published in the Journal of Finance in 1999. It covers 39 markets from the 1920s, and it measures capital appreciation only, with no dividends. Their abstract puts it plainly: "For 1921 to 1996, U.S. equities had the highest real return of all countries, at 4.3 percent, versus a median of 0.8 percent for other countries."
No country beat the US over the full period. The median of all 39 was 0.75% a year. Sweden came closest at 4.29%, and the authors note it also avoided major upheaval.
Two features of the design do most of that work. The first is the start date. Beginning in 1921 misses the 1914 to 1918 collapse in continental Europe, where German shares fell 21.67% a year in real terms while the exchange was shut. The second is dividends. Jorion and Goetzmann tested that objection directly and it doesn't rescue the other markets. Over 1970 to 1995 the income component was worth 4.14 percentage points a year in the US against a group average of 4.25, so the US was not unusually stingy. With dividends added back over 1921 to 1995, the US still led at 8.22% a year.
What changes the ranking is the measurement window, not the dividend treatment. That is worth holding onto whenever you see stock market returns by country quoted to one decimal place: the decimal is real, and it is also an artefact of where someone started counting.
The markets that died are the ones you can't see
Every table above is a table of survivors. In 1900 stock exchanges already existed in at least 33 of today's nations, and the DMS database covers 17 of the big ones with full histories. Russia's 1917 revolution and China's 1949 revolution both ended in expropriation, and Dimson, Marsh and Staunton record that both "experienced a compound rate of return of -100%." Russian investors in aggregate lost more than 99% in present-value terms even after the small bondholder settlements of the 1980s and 1990s.
Closure itself is the giveaway. Jorion and Goetzmann counted 24 markets with data in 1931. Seven traded without interruption: the US, Canada, the UK, Australia, New Zealand, Sweden and Switzerland. Seven were suspended for under a year, and the remaining 10 suffered a long-term closure. Market failure, in their words, "is not a remote possibility." In their table of breaks, Greece in October 1940, Romania in July 1941, Czechoslovakia in July 1943 and Hungary in July 1944 all have a blank where the restart date should be.
And closures aren't random. Jordà, Knoll, Kuvshinov, Schularick and Taylor make the point sharply in their 2017 Federal Reserve Bank of San Francisco paper: "Stock markets tend to be closed precisely at times when we would expect returns to be low, such as periods of war and civil unrest. Return estimates that exclude such rare disasters from the data will thus overstate stock returns." They then estimated the returns anyway. During the Portuguese revolution of 1974 to 1977, when the exchange was shut, real equity returns averaged -54.98% a year, a cumulative -90.88%. During the Spanish Civil War, with the exchange shut from 1936 to 1940, real returns averaged -4.01% a year. For Germany over 1914 to 1918, -21.67% a year.
This is the same selection problem that survivorship bias creates in fund league tables, scaled up to whole countries.
The counter-case: survivorship is worth about 0.1 percentage points
Here is where the evidence cuts against the intuition, and the people making the argument are the ones who built the surviving database.
Dimson, Marsh and Staunton addressed it head-on in their 2006 paper on the worldwide equity premium. Their estimate is that their 17 countries accounted for about 90% of world equity capitalisation in 1900, with omitted markets making up roughly 10%. Russia was the largest absentee at just under 5%; Austria-Hungary about 2%; China, on Goetzmann, Ukhov and Zhu's data, 0.4%.
Now assume the worst: every omitted market went to zero, and did so in a single catastrophic year. Spread a 10% loss of capitalisation across 106 years and the adjustment factor is 0.9 to the power of 1/106, which is 0.999. Their conclusion: "at most, survivorship bias could give rise to an overstatement of the geometric mean risk premium on the world equity index by about one-tenth of a percentage point." Jordà and his co-authors reach a near-identical figure from the other direction, putting the effect of market closures on average cross-country returns at "around -0.1 percentage points."
There's a second reason the effect is small, and it's easy to miss. The DMS world index already contains Russia and China at their true outcome. The Yearbook states that Chinese returns from 1900 and Russian returns are both incorporated into the world and world ex-US indices, with the communist takeovers treated as total losses for domestic investors. The total-loss cases most often cited as missing are, in that index, present. The Yearbook covered 98% of the world equity market in 1900 and 91% of it at the end of 2014.
So the two camps are not really disagreeing about arithmetic. Jorion and Goetzmann are right that single-country US data overstates what a 1900 investor could have expected. DMS are right that a capitalisation-weighted world index, which is what a global tracker approximates, barely moves when you add the dead markets back.
A second dataset, 16 countries back to 1870, ranks them differently again
The Jordà, Knoll, Kuvshinov, Schularick and Taylor database runs from 1870 and covers 16 advanced economies. Its headline table of average annual real equity returns puts Finland at 9.98%, the US at 8.39%, Sweden at 7.98% and Australia at 7.81%, with France at 3.25% and Portugal at 4.37% at the bottom. The authors observe that long-run risky asset returns in most countries sit "close to 6%-8% per year."
Those numbers are higher than the Yearbook's because they are arithmetic averages of annual returns rather than compounded rates, and the authors note the arithmetic average "is always larger than the geometric period-average return". The start date is 1870 rather than 1900 too. The ordering differs too. Finland is middling in the Yearbook and top here. The US is third there and second here. France is near the bottom in both, which the authors attribute to the world wars and the post-1945 wave of nationalisations.
Two independent teams, overlapping countries, different methods, materially different rankings. What survives both is the shape: a spread of six or seven percentage points a year between the best and worst national equity markets, and a bottom populated by countries that lost a war, a currency, or both.
One country is now 73% of a global tracker
The practical end of this is not the 1900 table but the 2026 one. On 30 September 2026 the MSCI World index, which covers 23 developed markets, weighted the United States at 72.94%, Japan at 5.91%, the United Kingdom at 3.41%, Canada at 3.29% and France at 2.24%. The UBS Global Investment Returns Yearbook 2026 puts the US at around 62% of total world equity market value, on a wider measure that reaches beyond developed markets.
A globally diversified equity holding is, by construction, mostly one country's record. Whether that's a problem depends on an assumption nobody can test: that the next hundred years ranks countries the way the last one did. The home bias question and the UK versus global equities question both hang on it.
Diversification across asset classes has done more reliable work than diversification across countries. The 2026 Yearbook notes that since 1900 equities and bonds have each "on several occasions lost more than 70% in real terms", while a 60:40 equity:bond blend "has never declined more than 50%". And equities still beat bonds, bills and inflation in all 21 Yearbook countries with continuous investment histories, which is the one result that held everywhere.
What this evidence cannot tell you
These are backtests of single, unrepeatable paths. Each country has one history, so a 115-year "average" has a sample size of one in the dimension that matters. Dimson, Marsh and Staunton say it themselves about the dispersion in their data: "just as some countries were 'lucky' and others less fortunate, some industries prospered while others foundered. And just as we can infer little about future country returns from past returns, we can infer little about long-run future industry returns from their historical record."
The construction choices are not neutral either. Deflating by a wholesale price index rather than consumer prices, measuring in local currency rather than dollars, starting in 1870 or 1900 or 1921, including or excluding dividends: each decision moves individual countries by whole percentage points and reshuffles the ranking. Stock market returns by country are a construction as much as a measurement, and the construction shows up in the ranking. There is no neutral version of this table.
Coverage thins as you go back. The Yearbook's early decades rest on reconstructed indices, often from a handful of large listed companies, and the authors are explicit that aggregate market-capitalisation data for 1900 were "neither recorded nor even thought of" at the time. The 10% omitted-capitalisation estimate is itself built on Rajan and Zingales's 1913 ratios plus, in the authors' words, "some informed guesses."
And none of it covers the catastrophes that didn't happen. Barro and Ursúa's long-term data for 30 countries up to 2006 found 232 stock-market crashes, defined as multi-year real returns of -25% or worse. Realised history is one draw from a distribution that contained worse draws.
What would change the conclusion
Three things would.
If the omitted markets were larger than 10%. The whole survivorship rebuttal rests on that single estimate, and it is an estimate, built from 1913 capitalisation ratios and informed guesses. If the true figure for non-surviving markets were 30% rather than 10%, the same formula multiplies the adjustment by about three and a half, and the Jorion-Goetzmann reading looks much stronger.
If a large modern market suffered a total loss. Nothing in the post-1950 record has tested a developed market to destruction. The frequency of failure in the first half of the twentieth century, 10 long closures out of 24 markets in 1931, is not the frequency observed since. Whether that's progress or a quiet accumulation of tail risk is unresolved, and a single expropriation in a major index market would settle it the hard way.
If US exceptionalism reversed. The US was third of 21 from 1900 to 2014 and is now roughly three-quarters of a developed-market tracker. Those two facts are in tension. A long stretch of US underperformance would not break the historical table, but it would end the convenient situation in which the best-documented market and the dominant market are the same market.
The thing worth watching isn't the long-run average at all. It's the gap between the country weights you hold today and the country weights that produced the average you're quoting. For most global investors those two numbers have never been further apart, and the long-run record on historical returns by asset class offers no view at all on which way that gap closes.