The Size Factor

The size factor is the observation that small companies have, on average, out-returned large companies by more than their market risk alone can explain. Banz spotted it in 1981, and it became one of the two original tilts in the The Fama-French Factor Models three-factor model, where it lives as SMB — "small minus big." The idea has intuitive appeal: small firms are riskier, harder to trade, and less followed, so they ought to pay you more. The reality is messier. Of all the classic factors, size is the shakiest, and much of a quant's job here is knowing why to distrust it.

The signal

SMB is built exactly like the value factor in the The Fama-French Factor Models 2×3 sort: split the universe into small and big by median market capitalisation, and take the return of the small-cap portfolios minus the big-cap portfolios, averaged across value buckets so it is (roughly) value-neutral.

$$\text{SMB} = \bar r_{\text{small}} - \bar r_{\text{big}} .$$

Here $\bar r_{\text{small}}$ is the average return of the small-cap portfolios and $\bar r_{\text{big}}$ that of the large-caps. A stock's exposure to SMB — its loading in a regression — tells you how much of its return is really just a small-cap tilt rather than skill.

Worked example: where the premium hides

Sort a universe into five size buckets and look at long-run average annual returns:

Size bucket	Avg. annual return	Notes
Micro-cap (smallest)	13.0%	tiny, illiquid, hard to trade
Small	10.5%
Mid	9.8%
Large	9.3%
Mega-cap (biggest)	9.0%

Naively, SMB looks like $13.0\% - 9.0\% = 4.0\%$ per year — a healthy premium. But watch what happens when you drop the micro-cap bucket, the one you can barely trade: the gap collapses to $10.5\% - 9.0\% = 1.5\%$, and after realistic trading costs in those tiny names, close to zero. Almost the entire "size premium" was concentrated in the smallest, least investable stocks — and much of that was inflated by survivorship and delisting gaps in old small-cap data.

Why size is so fragile

Three problems keep undermining SMB:

• It lives in the micro-caps. As the example shows, the premium is concentrated in stocks too small and illiquid to trade at scale, so it overlaps heavily with the liquidity premium — you may just be getting paid to hold what nobody can trade.
• It is seasonal and time-varying. A large chunk historically showed up in January alone, and the effect all but disappeared for long stretches after it was published in the 1980s — a classic decay warning sign.
• Data traps. Small-cap datasets are riddled with survivorship and delisting problems; failed tiny firms drop out, flattering the survivors' returns.

The "control your junk" rescue

The factor got a partial rehabilitation. Asness and co-authors showed that the raw size premium is weak because small-cap baskets are stuffed with "junk" — unprofitable, high-risk, low-quality firms that drag returns down. When you control for quality — compare small good firms to large good firms — a robust and stable size premium reappears. The slogan is literally the paper's title: size matters, if you control your junk. In practice this means size is best used conditionally, paired with a quality screen, rather than as a naked small-cap tilt.

Where it stumbles

• Illiquidity in disguise. Much of the premium is liquidity compensation you cannot capture net of the costs of trading tiny names.
• Capacity limits. Small-cap strategies fill up fast; a big book moves the very prices it trades, so paper returns overstate live ones (Portfolio Capacity, Market Impact).
• Post-publication decay. Standalone SMB has been near-zero for long spells since the 1980s — a caution that a well-known "premium" may already be arbitraged away.
• Definition sensitivity. Where you draw the size cutoffs and whether you value-weight changes the answer, so replication demands the exact recipe.