Key takeaways
- The 60-month rolling correlation between North American and developed ex-US equities peaked at 0.93 in the window ending February 2013, and read 0.80 in the window ending July 2026.
- Over the 433 months from July 1990 to July 2026, North America compounded at 10.93% a year in dollars against 6.33% for developed markets outside it.
- Holding 70% North America and 30% developed ex-US ran at 14.74% annualised volatility against 15.14% for North America alone, a saving of 0.40 points.
- Emerging markets are the exception. Their rolling correlation with North America fell from 0.89 in the window ending February 2013 to 0.68 by July 2026.
- From 2000 to 2009 North American equities returned 6.4% in total while emerging markets returned 168.8%, the decade the return gap ran the other way.
The short answer: correlations stopped rising in 2013, and the return gap did the damage
You hold a global equity fund. It's trailed a US index fund for most of a decade, and you want to know whether the foreign half is still doing anything for you.
Two things are true at once, and they point in opposite directions. Developed-market correlations did converge, and they converged fast, mostly between 1996 and 2005. Then they stopped.
On monthly dollar returns from Kenneth French's data library, the 60-month rolling correlation between North American and developed ex-US equities peaked at 0.934 in the window ending February 2013. In the window ending July 2026 it reads 0.804. That's thirteen years with no further convergence, which isn't what the convergence story predicts.
The part that actually hurt was never the correlation. Over the full 433 months to July 2026, North America compounded at 10.93% a year in dollars and the rest of the developed world at 6.33%. That 4.60-point gap is what a globally diversified investor gave up, and it has nothing to do with how closely the two moved together.
What the test uses, and why it starts in 1990
Every figure computed here comes from the regional return files in Kenneth French's data library at Dartmouth, in the version built from the July 2026 Bloomberg database. Each file carries a monthly value-weighted market return for one region, quoted in US dollars, with the risk-free rate reported alongside it. Six regions are used: North America, developed ex-US, Europe, Japan, Asia Pacific ex Japan, and emerging markets.
The developed regional series begin in July 1990, so that's the common start date, and July 2026 is the last month published. That gives 433 monthly observations for each region, and every correlation, volatility and return figure below is computed on them.
Two features of that data shape everything that follows. The returns are unhedged, so currency movements sit inside them, which is a separate lever with its own evidence on currency hedging. And they are regional buckets, not countries, so dispersion inside Europe or inside emerging markets is averaged away.
Global equity correlation rose once, in the 1990s, and then flattened
Split the sample by decade and the shape is hard to miss. The correlation between North American and developed ex-US monthly returns was 0.50 across the 1990s, 0.89 across the 2000s, 0.86 across the 2010s and 0.86 from January 2020 to July 2026.
The jump happened between the first decade and the second. After that, nothing moved. Three consecutive periods within 0.03 of each other is a plateau, not a trend.
The chart plots the same series at higher resolution, taking the 60-month rolling correlation at eight dates. It reads 0.39 in December 1995, 0.74 in December 2000, 0.86 in December 2005 and 0.93 at the February 2013 peak. Then it goes 0.88, 0.89, 0.81 and 0.80. The line rises for seventeen years and drifts sideways or down for the next thirteen.
Vanguard's April 2021 research reached the same reading from a different dataset. Its 10-year correlation between US and non-US stocks rose from 0.51 at 31 December 1989 to 0.86 at 30 September 2020, and the paper's own gloss is careful: "Since then, the trend in long-term correlations across equity markets has begun to flatten, perhaps implying that a ceiling to correlations among equity markets has been reached."
The academic record was never unanimous on the trend either. Bekaert, Hodrick and Zhang studied weekly returns for 23 developed markets from January 1980 to December 2003, 1,253 weekly observations, and reported that "we do not find evidence for an upward trend in return correlations" outside the European markets. That finding is contested, and the paper below disagrees with it, but a claim that half the literature can't reproduce isn't the settled fact it gets quoted as.
The correlation matrix for five equity regions, 1990 to 2026
Full-sample monthly correlations, July 1990 to July 2026, all in dollars:
- North America against Europe 0.81, developed ex-US 0.78, Asia Pacific ex Japan 0.73, emerging markets 0.73, Japan 0.48.
- Developed ex-US against Europe 0.93, Japan 0.79, Asia Pacific ex Japan 0.79, emerging markets 0.77.
- Europe against Asia Pacific ex Japan 0.77, emerging markets 0.75, Japan 0.56.
- Japan against Asia Pacific ex Japan 0.51, emerging markets 0.50.
- Asia Pacific ex Japan against emerging markets 0.87.
The flat claim that everything now moves together doesn't survive that grid. Japan's correlation with North America across 36 years is 0.48. Europe's is 0.81. The spread between those two is wider than the entire rise in the headline number since 2000, and it sits inside the same index most people own as one line.
The developed ex-US bucket looks tighter than its parts because it's dominated by Europe, with which it correlates 0.93. What you hold as "international" is mostly a European position wearing a global label, which is worth knowing before you decide how much home bias you're carrying.
What international diversification does to portfolio volatility, in points
Correlation is a proxy. The thing it's a proxy for is portfolio volatility, so measure that instead.
Over the full 433 months, North America alone ran at 15.14% annualised volatility. Developed ex-US alone ran at 16.32%, which is higher, not lower. Mix them 70/30 and the result is 14.74%.
So the measured diversification benefit across 36 years was 0.40 percentage points of annualised volatility. A 50/50 mix gives 14.84%, marginally worse than 70/30, because the foreign leg is the more volatile of the two and past a point its own volatility outweighs what the imperfect correlation saves.
By decade it gets smaller and less reliable. Across the 2010s, North America alone ran at 13.04% and the 70/30 mix at 12.96%, a saving of 0.08 points, which rounds to nothing. From January 2020 to July 2026 it was 17.58% against 16.82%, a saving of 0.76 points. Same rule, same two assets, an order of magnitude between the two answers.
That's the honest size of the risk gain from international diversification: a fraction of a volatility point, unstable across periods, sometimes indistinguishable from zero. It's a real number and it's a small one, and it sits in the same range as the rebalancing bonus, which most people also assume is bigger than it is.
The complaint isn't correlation, it's the 4.60-point return gap
Here's the number people are actually reacting to. Across the 433 months, a dollar in North America became $42.20. A dollar in developed markets outside North America became $9.16.
MSCI's own factsheets tell it in their numbers. At 31 August 2026 the MSCI EAFE Index had returned 10.07% a year over ten years against 13.56% for the MSCI World Index, and 6.58% a year since 31 December 1987 against 9.08%.
Note what those figures do not say. They don't say correlations rose, and they don't say diversification stopped reducing risk. They say one region compounded faster than another for a long time, which is a statement about returns, not about co-movement. Conflating the two is the central error in most writing on US vs international stocks.
Rolling ten-year windows show how persistent it's been. Of the 314 ten-year windows in the sample, developed ex-US beat North America in 83 of them, or 26%. In the window ending July 2026, North America is ahead by 4.92 points a year. In the window ending October 2000 it was ahead by 12.40 points a year, the widest in the sample, and that was the moment immediately before the gap reversed.
The decade when international diversification paid, and nobody remembers it
From January 2000 to December 2009, North American equities returned 6.4%. Not per year: in total, across the whole ten years. Over the same decade Europe returned 40.1%, developed ex-US 29.6% and emerging markets 168.8%.
That's the shape of the thing. The foreign allocation looked like dead weight through the late 1990s, then carried the portfolio for a decade, then looked like dead weight again through the 2010s, when North America returned 228.9% against 84.0% for developed ex-US.
Japan is the standing reminder that this cuts both ways. It returned minus 31.3% across the 2000s and compounded at 3.21% a year across the full 36 years, the worst of the six regions, while showing the lowest correlation with North America of any of them. Low correlation is not a synonym for good.
The recent data has turned again. In calendar 2025 North America returned 18.3% and developed ex-US 31.9%. MSCI's EAFE Index returned 31.89% gross in 2025 against 21.60% for MSCI World. One year settles nothing at all, and the 10-year gap is still wide, but the direction is worth recording rather than assuming away.
The strongest case against: markets crash together, and the research agrees
The serious objection isn't that average correlations are high. It's that they rise exactly when you need them low.
Christoffersen, Errunza, Jacobs and Langlois put that to the test in the Review of Financial Studies in 2012, using weekly returns from January 1973 to June 2009 and a dynamic copula model, which is a way of measuring joint behaviour in the tails rather than on average. Average copula correlation across 16 developed markets rose from roughly 0.2 in the mid-1970s to around 0.8 in 2009. Their verdict was blunt: "whereas diversification benefits have largely disappeared for DMs, EMs still offer substantial diversification benefits."
French's data supports the first half of that. Take the worst 43 months for North America, the bottom decile of the sample, where the average North American return was minus 7.80%. In those same months developed ex-US averaged minus 7.42% and emerging markets minus 8.35%. Foreign equity didn't cushion the fall. It fell alongside.
Index drawdowns say it more starkly. MSCI EAFE's maximum drawdown was 60.15%, between 31 October 2007 and 9 March 2009. MSCI World's was 57.46%, over exactly the same dates. The diversified index and the foreign-only index bottomed on the same day.
One statistical caution belongs here. The measured correlation inside that worst decile is 0.68, below the 0.78 full-sample figure, and that gap is an artefact of conditioning on extreme moves rather than evidence of protection. The mechanism, and why a crisis correlation read off a truncated sample misleads in both directions, is the subject of our piece on correlation instability. The averages above avoid the problem by reporting returns rather than a conditional correlation.
The counter to the counter comes from Clifford Asness, Roni Israelov and John Liew, writing in the Financial Analysts Journal in 2011 under the title "International Diversification Works (Eventually)". Across 22 countries and monthly data from January 1950 to December 2008, the average worst month for a local portfolio was minus 27.0% and for a global portfolio minus 23.3%. At one month, diversification barely helped, and global portfolios were more negatively skewed than local ones.
Stretch the horizon and it inverts. At five years, the average worst local return was minus 57%, against minus 39% for the average worst global return. The extra negative skew fades too: "as we pass five years, even the point estimates no longer differ." Their case is that crash correlation is the wrong test, because a long, grinding domestic bear market does more damage to a lifetime portfolio than a shared three-month panic. What worked in each specific crash is a different question, covered separately in our study of diversification in a crisis.
Where this evidence runs out
Five limits, stated plainly.
- Everything is in US dollars and unhedged, so a sterling or euro investor sees different numbers. The correlations move less than the returns do, but they do move.
- The sample is one 36-year stretch beginning in July 1990, which starts after Japan's peak and contains exactly one global financial crisis. That's a small number of independent events.
- Monthly returns say nothing about what happens inside a week, and crisis co-movement is often a daily phenomenon.
- Regional buckets average away country dispersion, which is where a lot of the remaining benefit plausibly lives.
- Every figure here is a backtest. It describes what did happen, and no part of it forecasts what happens next.
What would change the conclusion
The convergence claim gets a clean rematch in the rolling correlation itself. If the 60-month figure climbs back above 0.934 and holds there for several years, then the flat stretch since 2013 was a pause rather than a ceiling, and Vanguard's guess about a ceiling was wrong.
Emerging markets are the second thing to watch, and the more informative one. Their rolling correlation with North America has fallen from 0.892 to 0.679 since early 2013, which is the single largest divergence in this data. If that reverses and emerging markets settle at developed-market levels, the last structural argument here weakens considerably.
The return gap is the third, and it's the one that decides how the question feels rather than how it's answered. Developed ex-US has won 26% of rolling ten-year windows. If that share keeps falling through a fourth decade, the case for holding foreign equity rests entirely on a 0.40-point volatility saving, and that's a thinner argument than most global funds are sold on.
What would not change it is another year of US outperformance. The 2000s are already in this sample, and so are the 2010s, and the two decades disagree completely. A test that flips its answer every ten years is telling you something about the horizon of the test, not about whether diversifying across countries works.