The Disappearance of the Size Effect After 1981

Rolf Banz's 1981 study of NYSE stocks going back to 1926 found something that didn't fit the standard asset-pricing models of the time: after adjusting for market risk (beta), stocks with the smallest market capitalizations had delivered noticeably higher average returns than stocks with the largest market capitalizations, decade after decade. The gap was large enough — several percentage points a year — that it couldn't easily be waved away as noise, and it didn't fit CAPM's prediction that only market risk should matter for expected return.

The paper landed at a pivotal moment: index-fund investing was growing, quantitative factor research was becoming a legitimate academic and practitioner field, and Banz's finding was one of the first clear "anomalies" — a documented, persistent pattern the dominant pricing model couldn't explain. Small-cap mutual funds and index products proliferated through the 1980s specifically to capture it.

What actually happened after 1981

The size premium weakened substantially in the years following publication, and by the late 1990s and 2000s, many studies found it had become inconsistent or vanished in large stretches of US data, particularly once returns were measured excluding the very smallest, least liquid "micro-cap" stocks that are expensive and difficult to actually trade at the prices used in academic databases. Three explanations dominate the debate. First, capital flows: billions of dollars moved into small-cap strategies after the paper's publication, plausibly bidding away part of the premium — a version of the same crowding dynamic that hit stat arb and merger arb. Second, liquidity and data issues: much of the historical size premium was concentrated in January and in the smallest, most illiquid deciles, raising the possibility that reported returns overstated what was actually achievable once realistic transaction costs and stale prices were accounted for. Third, some researchers argue the effect never fully died, but instead got absorbed into other, better-specified factors (like a distinct "quality" or "profitability" factor) once those were separated out statistically.

Worked example

Suppose in the pre-1981 sample, the smallest-decile stocks beat the largest-decile stocks by an average of 6 percentage points a year, risk-adjusted. A fund launches in 1983 specifically to harvest that premium, indexing to the bottom two deciles by market cap. Over the next twenty years, the realized premium the fund actually captures — after real trading costs on genuinely illiquid micro-caps, and after accounting for the fact that "small cap" as an investable category became far more crowded — comes in closer to 1–2 points a year, and in some sub-periods is negative. The published number and the tradeable number diverged, both because of capital flows into the strategy and because paper backtests understated real-world execution costs on the smallest names.

What this means in practice

The size effect's fade is now a standard cautionary example taught alongside the Value Line anomaly and momentum crowding: a documented risk premium can be real in a historical sample and still be a poor forward-looking trading strategy, once you account for the effect of the discovery itself on future capital flows, and for the gap between backtested and tradeable returns in illiquid names.