Key takeaways
- Over the 100 years from July 1926 to June 2026, the smallest 30% of US listed companies returned 11.75% a year against 10.27% for the largest 30%, a gap of 1.48 percentage points.
- Small beat large in 66.8% of the 1,081 rolling 10-year windows in that record, in 50.7% of one-year windows, and in 82.3% of 20-year windows.
- Strip out January 1975 to June 1983 and the advantage across the remaining 91.5 years falls to 0.05 points a year, from 9.79% against 9.74%.
- The Russell 2000 returned 10.64% a year over the decade to 31 July 2026 against 14.82% for the Russell 1000, at 20.72% volatility against 15.56%.
- Measured against large caps, the small bucket fell 71.5% from July 1983 to March 1999 and sat 50.4% below that 1983 peak in June 2026.
Small caps beat large caps two-thirds of the time, not always
"Small caps always beat large caps" is one of the oldest lines in investing. You've probably met it attached to a single number: the smallest US companies have compounded faster than the biggest ones for a century. That part is true. The word doing the damage is "always".
Here's the record. Kenneth French publishes monthly returns for US stocks sorted into size buckets at NYSE breakpoints, and the series runs from July 1926 to June 2026. Over those 100 years the bottom 30% by market value returned 11.75% a year. The top 30% returned 10.27%. The gap is 1.48 percentage points a year, and compounded over a century that is a very large number.
Now look at how it arrived. Across the 1,081 rolling 10-year windows in the sample, the small bucket beat the large bucket 66.8% of the time. Over one-year windows it was 50.7%, which is a coin flip. Over 20-year windows it was 82.3%. So the defensible version of the claim is that small caps beat large caps more often than not, over long holding periods, with a third of decades going the other way.
That is a different claim from the one people repeat. It's also the one the rest of this piece tests.
One 8.5-year stretch carries the whole century
The 1.48-point gap isn't spread evenly across the record. It's barely spread at all.
Between January 1975 and June 1983 the small bucket returned 35.15% a year. The large bucket returned 16.13%. That is 19.02 points a year for 8.5 years, and it's the most extreme stretch anywhere in the series.
Take that stretch out and recompute the remaining 91.5 years. The small bucket returns 9.79% a year. The large bucket returns 9.74%. The gap is 0.05 points, which is nothing.
None of that says the 1975 run was fake. It happened, and anyone holding small companies through it was paid handsomely. It says something about what kind of claim the century-long average supports. An advantage that survives only because of one 8.5-year window is a fact about that window as much as about small companies.
Four of the ten decades went to large caps
The chart below plots the small bucket's annualised return minus the large bucket's, decade by decade, in percentage points a year.
Small caps took the 1930s, 1940s, 1960s, 1970s and 2000s, and the 1950s were a near draw at 0.52 points. Large caps took the 1980s by 3.67 points a year, the 1990s by 4.54, the 2010s by 1.97, and the 2020s so far by 6.47 points through December 2025. The biggest single decade for small caps was the 1940s at 9.90 points a year.
Four losing decades out of ten is not a footnote. Someone who bought the small bucket at its July 1983 high and held for 30 years spent every one of the next 360 months behind where they started, and the same pattern shows up in equal weight vs market cap, where a size tilt arrives by a different route.
The 1983 peak that has never been recovered
Calendar decades are an arbitrary slicing. The more honest view tracks the ratio of the two buckets' cumulative wealth, because that shows when one is gaining on the other and when it's giving ground back.
That ratio peaked in July 1983. It then fell 71.5% to March 1999. Between August 1983 and March 1999 the small bucket returned 8.67% a year while the large bucket returned 17.73%, a gap of 9.07 points a year sustained for 15.7 years.
It hasn't recovered. In June 2026 the small bucket is still ahead over the full century, but its lead sits 50.4% below the July 1983 peak, 42.9 years later. That is longer than most people's entire investing life.
The worst single 10-year window ran from April 1989 to March 1999, when large caps won by 8.49 points a year. And the losing run was systematic rather than a blip. Every one of 148 consecutive monthly start dates, from February 1979 to May 1991, produced a 10-year window that large caps won.
What you could actually buy: Russell 2000 against Russell 1000
French's buckets aren't investable, and that matters more than it sounds. In June 2026 the bottom 30% held 1,864 companies carrying 1.45% of the aggregate market value of the whole sort. The top 30% held 518 companies carrying 92.02%. The bottom decile was 1,242 companies holding 0.35%.
So the academic small-cap record is largely a record of a sliver of the market, priced in stocks most funds are too big to own. That is the gap between the factor and the fund, and it's the same gap that shapes small cap tilt sizing in a real portfolio.
Index data closes the gap. On FTSE Russell's own factsheets, dated 31 July 2026, the Russell 2000 returned 10.64% a year over 10 years and the Russell 1000 returned 14.82%. That's 4.18 points a year to large caps, over a full decade, in the two indices people actually hold. The Russell 2000's median constituent was worth $1.146 billion; the Russell 1000's median was $17.273 billion.
The risk side reads worse than the return side. The Russell 2000's 10-year annualised standard deviation was 20.72% against 15.56% for the Russell 1000. Their Sharpe ratios were 0.48 and 0.82. More volatility, less return, and a bit over half the reward per unit of risk. In 2024 alone the Russell 2000 returned 11.54% against 24.51%. Shorter windows read the same way: 7.11% a year against 12.12% over five years, and 15.09% against 18.98% over three.
The most recent year points the other way. In the 12 months to 31 July 2026 the Russell 2000 returned 34.18% against 18.94% for the Russell 1000. One year is not evidence of a turn, which is the whole reason for the 1,081-window count above.
The record outside the US, and the UK version of it
US data isn't world data, so here's the rest of it. Small cap underperformance over the past decade shows up well outside the US. MSCI's factsheets, both dated 31 July 2026, put the MSCI World Small Cap index at 10.50% a year over 10 years in gross dollar terms, against 13.29% for MSCI World. That's 2.79 points a year to large and mid caps across 23 developed markets. The risk gap travels with it. Over the same decade MSCI World Small Cap ran a 17.79% annualised standard deviation against 14.85% for MSCI World, at Sharpe ratios of 0.51 and 0.76. The breadth problem repeats too: the small-cap index carries 3,877 constituents covering about 14% of each country's free float, against 1,282 constituents and 85% for MSCI World.
The UK is starker still. On FTSE Russell's factsheets at 31 July 2026, the FTSE 100 returned 13.2% a year over five years. The FTSE 250 returned 4.1% and the FTSE SmallCap returned 5.5%. UK mid caps trailed UK blue chips by 9.1 points a year while running 15.2% volatility against the FTSE 100's 10.1%. A UK investor who tilted small over that stretch paid for it twice, in return and in variance.
The strongest objection: small-cap indices are full of junk
The serious counter-argument doesn't dispute any figure above. It disputes the test.
In "Size matters, if you control your junk", published in the Journal of Financial Economics in 2018, Clifford Asness, Andrea Frazzini, Ronen Israel, Tobias Moskowitz and Lasse Pedersen take the complaints against the size effect seriously. Their abstract lists eight of them. The size premium, they write, has been accused of "having a weak historical record, being meager relative to other factors, varying significantly over time, weakening after its discovery, being concentrated among microcap stocks, residing predominantly in January, relying on price-based measures, and being weak internationally". Their answer is that small-cap indices are stuffed with unprofitable, indebted and volatile companies, and that those firms drag the average down. Strip them out, and "these challenges disappear when controlling for the quality, or its inverse, junk, of a firm".
They report the rescued premium holding across 30 industries and 24 international equity markets. That objection is well supported by their data, and it's the reason the debate is still live rather than settled. It also changes the claim being made. "Small caps beat large caps" becomes "small, profitable, low-debt companies beat large caps", which is a different portfolio and needs a screen to build. The Russell 2000 applies no such screen. Neither does French's bottom 30%. Anyone quoting the century-long number is quoting the unscreened version, and the unscreened version is the one that spent 15.7 years losing. For the post-publication half of the argument, there's more in the size premium after Banz.
What this sample cannot tell you
Three limitations, stated plainly.
First, this is one country's record for the long series. The French data covers NYSE, AMEX and NASDAQ stocks only, and the 100-year sample contains one Depression, one post-war boom and one technology bubble. Those are single events, not repeatable draws.
Second, the returns are gross. No trading cost, no spread, no tax, no fund fee. Small companies cost more to trade than large ones, so the realised gap has been narrower than the measured one, by an amount this data cannot tell you.
Third, rolling windows overlap heavily, so 1,081 ten-year windows aren't 1,081 independent observations. The 66.8% is a description of one path through history. It isn't a probability, and it doesn't say what the next decade holds.
The verdict on "small caps always beat large caps"
- Claim: small caps always beat large caps.
- Evidence for: 11.75% a year against 10.27% over the 100 years to June 2026, and a win in 82.3% of 20-year windows.
- Evidence against: large caps took four of the ten decades, the 15.7 years from August 1983, the 10 years to 31 July 2026 in the Russell and MSCI data, and the five years to that date in the FTSE data.
- Verdict: false as stated. The century-long average survives, but it rests on January 1975 to June 1983. Remove that and 91.5 years produce 0.05 points a year.
What would overturn this reading isn't another good year for small caps. It's the ratio. If small-to-large cumulative wealth regained its July 1983 level, the 15.7-year failure would stop being an open wound and become a completed cycle, and the claim would have survived its worst test on its own terms. In June 2026 that ratio sits 50.4% below the 1983 mark. That's the number worth watching, and it doesn't move quickly.