Key takeaways
- A 20% small cap tilt produced 3.14% annualised tracking error against the US market over the century to June 2026. A 5% tilt produced 0.78%, almost exactly a quarter of it.
- From 1927 to 2025 that 20% tilt added 0.46 percentage points a year. From 1981 to 2025 the same tilt cost 0.14 points a year, and from 2006 to 2025 it cost 0.53.
- Tilt size doesn't change the information ratio. At the full-sample ratio of 0.21, separating the size premium from zero at a t-statistic of 2 takes roughly 93 years of returns.
- MSCI's Standard indexes target 85% of free float market value, so a MSCI World tracker holds no small caps at all, while an ACWI IMI tracker already holds about 14%.
- At 31 July 2026 a global small cap tracker charged 0.45% a year against 0.12% for the developed-market fund, so a 20% tilt added 6.6 basis points of cost.
What a small cap tilt actually buys, and it isn't return
You've decided small companies deserve a place in your equity sleeve. The question left is how big: 5%, 10% or 20%. Here's the short answer. The only thing tilt size reliably delivers is tracking error, the year-to-year gap between your portfolio and the index you'd otherwise have held. That gap scales in a straight line with the size of the tilt, and you can price it before you start.
Using Kenneth French's size portfolios, the smallest 30% of US stocks by market value against the whole US market, a tilt funded out of the broad index produced these annualised tracking errors over the hundred years from July 1926 to June 2026:
- 5% tilt: 0.78% tracking error
- 10% tilt: 1.57%
- 20% tilt: 3.14%
Those numbers aren't estimates of anything uncertain. They're arithmetic, and the chart above shows how little the era changes them. Return is the part nobody can price in advance, and the record is far less flattering than the tracking error is predictable.
The arithmetic is fixed before the market opens
A tilt is a subtraction. If you hold the broad index and move a slice of weight into small caps, your portfolio's return differs from the index's by exactly that slice multiplied by the gap between small caps and the market. Put 10% into small caps and your active return every year is 10% of the small-minus-market difference. Nothing else in the portfolio moves.
Tracking error is the standard deviation of that active return, so it inherits the same multiplier. Over the full French sample, monthly small-minus-market differences annualise to 15.69% of tracking error at a full 100% swap. Multiply by 0.05, 0.10 or 0.20 and you have the three numbers above. Halve the tilt, halve the tracking error. There's no threshold where a tilt suddenly starts working, and no size at which it stops costing you the deviation.
This is why the phrase "moves the needle" is doing more work than it looks. It moves one needle, deterministically. The other needle, return, is not connected to the dial.
Ninety-nine years put a 20% tilt 0.46 points a year ahead
Over 1927 to 2025, rebalancing each year back to the target weight, the broad US market compounded at 10.27% a year. Adding the small cap sleeve moved that as follows:
- 5% tilt: 10.40% a year, an advantage of 0.12 percentage points
- 10% tilt: 10.51%, an advantage of 0.24 points
- 20% tilt: 10.73%, an advantage of 0.46 points
So the century says the tilt paid. It also says the payment was proportional, like the risk, which means the ratio between them never improved. A 20% tilt bought four times the tracking error of a 5% tilt and delivered a bit less than four times the return advantage.
Look inside the century and the smoothness disappears. The average annual gap between small caps and the market was 3.85 percentage points, with a standard deviation of 18.56 points. Forty-seven of the 99 years were negative. In 1929 small caps trailed the market by 32.07 points, which a 20% tilt would have converted into 6.41 points of underperformance in a single calendar year.
Since 1981 the same small cap tilt has been a drag
Rolf Banz published "The relationship between return and market value of common stocks" in the Journal of Financial Economics in 1981. The years since are the ones an investor today has actually lived through. Over 1981 to 2025 the average annual small-minus-market gap was minus 0.62 points. The tilted portfolios landed here:
- 5% tilt: 11.71% a year against 11.74% for the market, a shortfall of 0.03 points
- 10% tilt: 11.68%, a shortfall of 0.06 points
- 20% tilt: 11.60%, a shortfall of 0.14 points
Shorten the window again and it gets worse. Across 2006 to 2025 the average gap was minus 2.49 points a year, and a 20% tilt trailed by 0.53 points annually. On a starting balance of $10,000, that's a finish of $74,143 against $81,638 for the untilted portfolio, a difference of $7,494 over twenty years. Our piece on the size premium after Banz reaches the same place from the factor side.
The tracking error over that same post-1981 stretch was 0.57%, 1.15% and 2.29% at the three tilt sizes. Slightly lower than the century figure, because the 1930s are gone, and still perfectly proportional.
The information ratio doesn't care how big your tilt is
Divide the average active return by the tracking error and you get the information ratio, the return you earned for each unit of deviation you accepted. Because both halves scale with tilt size, the ratio is identical at 5%, 10% and 20%. Sizing is a volume knob, not a quality knob.
The volume matters less than what's coming through the speaker. Over the full 99 years the information ratio was 0.21, which sounds respectable until you ask how long you'd have to wait to be confident it isn't zero. The t-statistic on 99 years of that ratio is 2.06. Reaching a t-statistic of 2 at that ratio takes about 93 years of data, and we've had 99. Over 1981 to 2025 the ratio was minus 0.05, and over 2006 to 2025 minus 0.30.
That's the honest shape of the evidence. A century of the best size data available clears the conventional bar for statistical significance by a hair, and only if you're prepared to treat the 1930s as informative about the 2030s.
Your index fund may already hold 14% in small caps
Before sizing anything, it's worth knowing what a neutral weight looks like. MSCI's Global Investable Market Indexes methodology, in the May 2018 edition, sets market coverage targets of "85% ± 5%" for the Standard Index and "99% +1% or -0.5%" for the Investable Market Index. It then defines the small cap segment as the difference: "The Small Cap Index market coverage in each market is derived as the difference between the free float-adjusted market capitalization coverage of the Investable Market Index and the Standard Index in that market."
That difference is about 14%. The MSCI World Small Cap Index factsheet dated 31 July 2026 puts it plainly: 3,877 constituents covering "approximately 14% of the free float-adjusted market capitalization in each country". The MSCI World Index, the large and mid cap segment, had 1,282 constituents and about 85% coverage on the same date.
So the starting point isn't the same for everyone. A fund tracking MSCI World holds no small caps, and a 10% small cap allocation takes that holding from 0% to 10%, still below the market's own 14%. A fund tracking MSCI ACWI IMI, which State Street's tracker holds across 8,170 constituents, already carries roughly 14%. Adding 10% on top of that lands you at 22.6%, an overweight of 8.6 points. Same label, two different trades.
What you tilt into changes the tracking error more than how much
The 3.14% figure comes from US data, and the US small cap universe defined on NYSE breakpoints reaches a long way down. French's construction note is explicit: the portfolios "are constructed at the end of each June using the June market equity and NYSE breakpoints", covering "all NYSE, AMEX, and NASDAQ stocks" with market equity data. That bottom 30% includes microcaps no global tracker owns.
Run the same calendar years side by side and the difference is stark. Over 2012 to 2025, the annual small-minus-market gap on French's US data had a standard deviation of 7.94 points. For MSCI World Small Cap against MSCI World, both global and both broadly diversified, the standard deviation of the same 14 annual gaps was 4.60 points. A 20% tilt therefore bought 1.59% of annual tracking error in the US series and 0.92% in the global one.
Both were negative on average over that stretch: minus 3.48 points a year for the US series, minus 1.71 for the global pair. Over the ten years to 31 July 2026, MSCI World Small Cap returned 10.50% a year against 13.29% for MSCI World, a shortfall of 2.79 points, with annualised standard deviation of 17.79% against 15.21%.
The counter-case, given its best shot
There's a serious argument on the other side, and it doesn't depend on the recent record improving. It starts with the observation that a 99-year sample with a positive mean and a t-statistic above 2 is the strongest long-run evidence any equity style has, and that 45 years of disappointment inside it is not enough to overturn it. Value investors made the same argument about the value premium and the arithmetic is comparable.
The second strand is diversification rather than return. MSCI World Small Cap held 3,877 constituents at 31 July 2026 against 1,282 in MSCI World, whose top holding alone carried 5.18% of the index. A tilt toward the small end is a tilt away from that concentrated top.
The third is that the cost of being wrong is bounded and known in advance. At a 5% tilt the tracking error is 0.78%, less than the spread between a cheap and an expensive index fund. Our comparison of tracking difference vs tracking error puts that gap in context. What the counter-case cannot do is convert a bounded cost into an expected gain, and it doesn't claim to.
The fee gap and the volatility gap are both small, and both real
Tilting costs money in two places. The first is the fund fee. State Street's SPDR MSCI World Small Cap UCITS ETF carried a total expense ratio of 0.45% at 31 July 2026, against 0.12% for its MSCI World fund and 0.17% for the ACWI IMI version. The 0.33 point gap applies only to the tilted slice, so a 5% tilt added 1.7 basis points a year and a 20% tilt added 6.6.
The second is portfolio volatility, and it barely registers. Over 1981 to June 2026, annualised volatility on the untilted US portfolio was 15.45%. At a 5% tilt it was 15.55%, at 10% it was 15.67%, and at 20% it was 15.97%. Half a percentage point of extra volatility for the largest tilt tested.
That asymmetry is the point worth carrying away. A small cap tilt is not primarily a volatility decision, because total risk hardly moves. It's a tracking error decision, and tracking error is the risk of holding something different from what everyone else holds, for years, without knowing whether it's working.
What this evidence can't tell you
Start with geography. The French series is one country. The UBS Global Investment Returns Yearbook 2026 covers "all the main asset categories in 35 markets" and draws on "more than 125 years of historical data", and it finds developed markets delivered 8.5% annualised equity returns since 1900 against 6.9% for emerging markets. Country differences of that size are normal. A US size result is not a world size result, which is why the MSCI comparison sits above.
Second, none of this is a forecast. Every figure here is one realised path. The 1927 to 2025 advantage rests heavily on a handful of enormous years, 1933 above all, when small caps beat the market by 93.34 points.
Third, the tracking errors assume you hold the tilt through the bad stretches. The worst single year since 1981 was 1998, when small caps trailed by 29.96 points and a 20% tilt would have cost 5.99 points of relative return. A tilt abandoned in year four of a drawdown realises the tracking error and none of the premium.
Fourth, the returns above are gross of tax and of trading costs beyond the fund fee. Rebalancing a tilt in a taxable account triggers events the arithmetic here ignores. Our work on rebalancing band width covers what that maintenance actually involves.
What would change the conclusion
Three things would move it. The first is a break in the proportionality itself. If small cap returns became less correlated with the broad market, the same tilt would buy more tracking error per point of allocation, and the sizing arithmetic here would need redoing rather than rescaling.
The second is concentration at the top. The MSCI World top holding sat at 5.18% of the index at 31 July 2026. If that share keeps climbing, the diversification case for a tilt strengthens on its own terms, independent of whether the size premium ever reappears. If it falls back, that case weakens.
The third is the fee gap. At 0.33 points it's small enough that a 20% tilt costs 6.6 basis points. If small cap trackers converge on broad-market pricing, the bounded cost of a modest tilt shrinks toward the tracking error alone, and the tracking error is the part you can already calculate. LedgerTouch shows the drift between a target tilt and a live one, which is the number the arithmetic above turns on.