Fama-French Analytics

The Size Effect

The size effect is the observation that small companies have historically earned higher returns than large ones. It has the strangest history of any of the classic factors: it was discovered first, it was the most immediately intuitive, and it is the one that has held up worst under scrutiny. Understanding why is genuinely useful, because the ways the size premium was overstated are the same ways every other factor can be overstated.

Banz and the original finding

Rolf Banz documented the effect in 1981, showing that the smallest companies on the NYSE had earned returns well above what the Capital Asset Pricing Model predicted from their market exposure alone. This was one of the first serious cracks in the single-factor view. The finding was intuitive enough to be adopted quickly: small companies are more fragile, harder to trade, less researched, and more likely to fail, so it seemed reasonable that investors would demand extra return to own them.

When Fama and French built the three-factor model in 1993, size was the natural second factor alongside value. SMB, or Small Minus Big, is constructed by going long a portfolio of small-capitalisation stocks and short a portfolio of large ones, so like HML it measures a spread rather than a directional return.

SMB = Average return on small portfolios - Average return on big portfolios

What happened after publication

The size premium weakened substantially after Banz published, and the decline was sharp enough that it became the standard example of a factor that may have been arbitraged away once it was known. This pattern, where a documented anomaly shrinks after publication, has since been found across a wide range of published factors and is one of the more sobering regularities in the literature.

But publication effects are not the whole story, and several measurement problems inflated the original estimate:

  • Survivorship and delisting bias. Small companies fail far more often than large ones. Early databases handled delistings poorly, and a company that goes to zero can quietly vanish from a sample rather than being recorded as a total loss. This flatters small-cap returns specifically.
  • Illiquidity and trading costs. The smallest companies in the sorts are often the least liquid, with wide bid-ask spreads. A paper portfolio rebalances at the midpoint. A real one does not, and for micro-caps the gap between the two is large enough to consume a meaningful share of the premium.
  • Concentration in the extreme tail. Much of the measured effect came from the very smallest companies, which are also the ones least investable at scale. A fund of any size cannot own them in meaningful weight.
  • Seasonality. A disproportionate share of the historical premium showed up in January, a concentration that is hard to reconcile with a risk-based explanation and easier to attribute to tax-loss selling and related effects.

The general lesson: every one of these problems inflates a backtest without any dishonesty on the researcher's part. They are the default outcome of careful work on imperfect data, which is why the backtesting article in this section treats them as the rule rather than the exception.

Why size stayed in the model anyway

Given all of this, it is fair to ask why Fama and French kept size in the three-factor model, and kept it again in the five-factor model in 2015. The answer is that size earns its place as a control rather than as a premium to be harvested.

Value, profitability, and investment effects all behave differently among small companies than among large ones, and in several cases they are considerably stronger in the small-cap universe. Without a size factor in the model, those interactions contaminate the other estimates. Fama and French construct HML, RMW and CMA within size buckets for exactly this reason. Size is the variable that makes the other factors measurable cleanly, whether or not it pays a premium of its own.

This dashboard reflects that structure. The small-value minus small-growth and big-value minus big-growth spreads let you see whether the value effect is currently concentrated in smaller companies or spread evenly across the market, and the small-value minus big-value spread isolates the size dimension within value stocks. When these diverge sharply it usually means a broad statement about "value" is hiding two different stories.

What a careful investor takes from this

The defensible reading of the evidence is that a standalone small-cap tilt is a weak proposition, that the interaction between size and other factors is where the more durable evidence sits, and that any estimate of a small-cap premium taken from a backtest should be discounted for costs that the backtest almost certainly did not charge. That is a less exciting conclusion than the 1981 result, and it is better supported.

None of this constitutes investment advice. It is a description of what the historical data shows and how the profession's reading of it has changed.

Banz, R. W. (1981). "The relationship between return and market value of common stocks." Journal of Financial Economics, 9(1), 3-18.

Fama, E. F., & French, K. R. (1993). "Common risk factors in the returns on stocks and bonds." Journal of Financial Economics, 33(1), 3-56.