The Size Factor (Small-Minus-Big) — a textbook case of post-publication decay this wiki uses as a yardstick

Small firms beat large firms by ~0.4-0.5% per month in the original 1936-1975 sample (Banz 1981) — and that premium collapsed to statistically indistinguishable from zero after 1983. What survives is a microcap liquidity story, not a size story, and even that partial survival needs a quality control to resurface. This is the wiki's cleanest example of Out-of-Sample vs Post-Publication Decay: The Two Numbers That Tell You If a Premium Is Real: a real, peer-reviewed effect that mostly stopped working once it was published and arbitraged.

Banz 1981: the original effect

Banz analyzed all NYSE common stocks, 1936-1975, and found the coefficient on market value negative and significant — smaller firms earned higher risk-adjusted returns than CAPM predicted (https://www.researchgate.net/publication/228205782_A_Literature_Review_of_the_Size_Effect). The size:

- Smallest 20% of firms earned approximately 0.4% per month higher risk-adjusted return than large firms (annualized ~4.8%) — FONTE: https://www.researchgate.net/publication/228205782_A_Literature_Review_of_the_Size_Effect - On a nominal (non-risk-adjusted) basis, the average annual gain of the smallest firms was almost 20% higher than the largest ones — FONTE: https://fincyclopedia.net/exchanges/s-exchanges/small-firm-effect-2/ - The effect was read at the time as evidence that CAPM itself is misspecified, since size predicted returns after controlling for beta — FONTE: https://fincyclopedia.net/exchanges/s-exchanges/small-firm-effect-2/

Proof regime

original peer-reviewed study, Journal of Financial Economics 1981 — the first peer-reviewed documentation of the effect, and the founding case study for Proof Regimes: Peer-Reviewed vs Practitioner-Attested vs Unverified — how to grade the source of a claimed edge.

Post-1983 decay: the numbers that killed it

The decay is not a vague "it got weaker" — it is quantified precisely:

- Pre-1983 (1936-1983): smallest-vs-largest decile spread = 0.50% per month, t-statistic = 2.23 (highly significant) — FONTE: https://efmaefm.org/0efmameetings/efma%20annual%20meetings/2016-Switzerland/Papers/EFMA2016_0340_FullPaper.pdf (citing Dichev 1998, Chan et al. 2000) - Post-1983 (1983-present): same spread = 0.01% per month, t-statistic = 0.03 — statistically indistinguishable from zero — FONTE: same source - That is a 98% reduction in the size premium, from 0.50% to 0.01% per month - The U.S. small-firm effect was "strongly positive every year during 1974 to 1983, and then was negative for six out of the next seven years" (1984-1990) — FONTE: same source - 1978-2013: Russell 2000 (small-cap) returned 12.1% annually vs Russell 1000/S&P 500 (large-cap) at 12% — statistically flat over 35 years — FONTE: https://www.financierworldwide.com/a-quick-review-of-the-literature-regarding-the-small-cap-premium - Long-horizon estimates diverge sharply with the window chosen: Fama-French report 3.1%/year US small-cap outperformance for 1927-1981, while Jeremy Siegel reports only 1.8%/year for 1926-2012 — FONTE: same source - AQR's own test on 1963-2020 monthly data (unadjusted) finds decile-1-vs-decile-10 t-statistic "well south of 2.0" — weak significance; after adjusting for the higher market beta of small stocks, "the premium was cut roughly in half, with pretty much no size effect net of the larger betas of small stocks" — FONTE: https://www.aqr.com/Insights/Perspectives/There-is-No-Size-Effect-Daily-Edition

When it applies

any size-premium number quoted without specifying pre/post-1983 is not comparable to another — the two regimes differ by two orders of magnitude.

The liquidity/microcap problem, quantified

The residual premium, where it exists at all, is a liquidity story concentrated in illiquid microcaps rather than a broad size story:

- The characteristic liquidity premium was "large and robust until the mid-1980s but had become small and 'second-order' since" (citing Ben-Rephael et al. 2015) — FONTE: https://www.financierworldwide.com/a-quick-review-of-the-literature-regarding-the-small-cap-premium - Round-trip bid-ask spread cost: ~5 basis points for most liquid stocks vs 220 basis points for the least liquid microcaps — a ~44x differential — FONTE: https://canvas.osam.com/Commentary/BlogPost?Permalink=microcaps-factor-spreads-structural-biases-and-the-institutional-imperative - "Removing stocks with less than $5m in market cap causes the small firm effect to vanish" — indicating a microcap, not small-cap, phenomenon — FONTE: https://www.financierworldwide.com/a-quick-review-of-the-literature-regarding-the-small-cap-premium - Amihud (2002): the illiquidity effect is "stronger for smaller stocks, as the coefficient increases in magnitude as size decreases" — FONTE: https://papers.ssrn.com/sol3/Delivery.cfm/2451_26706.pdf?abstractid=1295244&mirid=1 - Outside the microcap universe, "virtually no liquidity effect can be observed" in international equities — FONTE: https://www.financierworldwide.com/a-quick-review-of-the-literature-regarding-the-small-cap-premium - ERISA effect: removing the 1975-1983 window (when U.S. pension law first allowed large funds to diversify into small caps) makes large- and small-cap returns converge to nearly identical 8% annual returns — FONTE: same source. This is the leading causal explanation offered for why the effect existed at all before it decayed. - Adjusting for illiquidity-induced beta underestimation with daily data (1-day and 2-10 day lags), small-stock alpha shifts from an apparent +2.3% to an adjusted -0.6% annually — FONTE: https://www.aqr.com/Insights/Perspectives/There-is-No-Size-Effect-Daily-Edition

Survives transaction costs

NO for realistic round-trip costs (50+ bps on microcaps) — the residual premium is absorbed by costs. This is a direct instance of Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one.

Conditional resurrection: quality changes the verdict

Controlling for firm quality reverses the unconditional post-1983 verdict:

- Asness, Frazzini, Pedersen ("Size Matters, if You Control Your Junk") find that once quality (QMJ) is controlled for, the size premium becomes "large and significant," stable across time, size measures, seasons, industries, and international markets — FONTE: https://www.researchgate.net/publication/254911501_New_Paradigm_or_Same_Old_Hype_in_Equity_Investing - Controlling for quality "more than doubles the average performance of the size factor and its significance," recovering the effect in the 1980s-1990s window where it was otherwise absent — FONTE: same source - The mechanism: small stocks have "very large negative exposure to quality" — they are "junky" relative to big stocks, so an unconditional size sort is partly a bet against quality — FONTE: same source - See Quality-Minus-Junk (QMJ): a Verified Multi-Decade, 24-Market Alpha — and Where Its Numbers Stop for the QMJ factor itself and its own numbers.

Quality is the load-bearing modification here — it inverts the original claim. It is quality, not size per se, that predicts returns in the post-1983 sample; "small-cap" alone is not a standalone edge.

What does NOT work

- Trading the unconditional, un-adjusted size premium post-1983: t-statistic ≈ 0.03, i.e. no detectable effect (EFMA 2016 paper above). - Trying to capture the residual premium via broad small-cap index products: the Russell 2000 vs Russell 1000 comparison (1978-2013) shows no meaningful gap (12.1% vs 12%). - Extracting the microcap liquidity premium net of realistic trading costs: costs (50+ bps round-trip) exceed the documented residual premium. - Treating any single-period size-premium number as portable across regimes: 1927-1981 (3.1%/yr, Fama-French) and 1926-2012 (1.8%/yr, Siegel) are not the same claim, and neither maps onto the 0.01%/month post-1983 figure.

Current academic consensus

The size effect has essentially disappeared in U.S. equities post-1983 when tested unconditionally (EFMA 2016 paper). The leading explanation is the 1975 ERISA legislative change that let pension funds diversify into small stocks, combined with later improvements in market liquidity (narrower spreads, lower commissions) that both created and then arbitraged away the mispricing — FONTE: https://www.financierworldwide.com/a-quick-review-of-the-literature-regarding-the-small-cap-premium. Dimensional's own retrospective calls it "a real, long-run statistical tendency backed by decades of evidence, not a reliable short- or medium-term timing signal," while conceding the evidence "is considerably more complicated than the summary statistic suggests" and that the premium has been "inconsistent, period-dependent, and possibly diminished since its discovery" — FONTE: https://www.dimensional.com/us-en/insights/the-evolution-of-small-cap-investing-four-decades-of-innovation-at-dimensional.

Proof regime & multiple-testing note

- Original discovery: peer-reviewed (Journal of Financial Economics 1981), in-sample on 1936-1975 data. - Post-publication: severe decay and out-of-sample failure — 0.50%/month pre-1983 to 0.01%/month post-1983. - Survives transaction costs: NO, for realistic round-trip costs on the microcap segment where any residual lives. - Survives out-of-sample/post-publication: PARTIALLY — only when controlling for quality (QMJ); the unconditional effect is essentially gone post-1983. - Multiple-testing concern: the size effect was among the first anomalies ever published, and the many subsequent reframings (quality-adjusted, conditional, by country) that were needed to keep it alive are themselves a flag for the kind of data-mining problem covered in The Factor Zoo and Finance's Replication Crisis — why most published factors are false discoveries and Multiple Testing: Why t>1.96 Is Not Enough — the bar this wiki uses to grade a factor's significance.

Why this page matters for the wiki

This is a worked example of the wiki's own evidence bar from What Counts as an Edge Here: The Evidence Bar This Wiki Applies to Every Technique: it has real numbers on all four required dimensions (effect size, proof regime, cost survival, out-of-sample persistence), and on three of those four dimensions the honest answer is "it failed." Anyone using this wiki to decide whether "small-cap" is an exploitable technique should read the 0.50%→0.01% collapse before the Banz 1981 headline number, and should read the QMJ-conditional result before assuming the whole line of research is dead — the size factor is neither a clean win nor a clean refutation, and pretending otherwise is exactly the kind of unsupported claim this wiki exists to filter out.