The Value Factor (Book-to-Market / HML) — a founding premium whose post-1991 numbers can't rule out zero

HML (High-Minus-Low book-to-market) is the value premium Fama and French formalized in 1992/1993: buy cheap-relative-to-book stocks, short expensive ones. Its original numbers were strong and long-lived; its post-1991 numbers are small enough that the paper's own authors say they cannot statistically distinguish "the premium shrank" from "the premium had 28 unlucky years." This page carries both sets of numbers and the debate over why.

Fama-French 1992/1993 original numbers and sample

- Large-cap value premium, July 1963 – June 1991: 0.36% monthly (~4.3% annualized). - Small-cap value premium, same period: 0.58% monthly (~7% annualized). - Sample: NYSE and AMEX stocks, 336 months (July 1963 – June 1991, 28 years). - Method: HML = mean return of two high book-to-market portfolios minus mean return of two low book-to-market portfolios, controlling for size. - Source for all of the above: Kenneth French interview reporting the original Fama-French figures (institutionalinvestor.com); the 1963–1991 large-value figure of 4.3%/year is corroborated independently by the Chicago Booth Review piece.

The premium shrinks after 1991 — numbers, not narrative

- Large-cap value premium, July 1991 – June 2019: 0.05% monthly (~0.6% annualized) — an 86% decline from the 0.36% baseline. - Small-cap value premium, same period: 0.33% monthly (~4% annualized) — a 43% decline from the 0.58% baseline. - Chicago Booth's framing of the same split: 1963–1991 large value averaged 4.3%/year; 1991–2019 it averaged 0.6%/year (Eugene Fama, chicagobooth.edu). - The catch, straight from the source that ran the numbers: across two 28-year sub-periods, the decline "seems large, but statistically [it is] indistinguishable from zero" — there isn't enough data to reject either "the premium genuinely fell" or "it just had a long unlucky stretch." Kenneth French's own verdict: "there is no way to tell."

When this matters for [[concept/what-counts-as-an-edge]]

a factor whose two half-sample means differ by 86% and still can't clear a significance bar is the textbook case for why effect size alone is not evidence — see Multiple Testing: Why t>1.96 Is Not Enough — the bar this wiki uses to grade a factor's significance for why the threshold has to be higher than the classic t>1.96.

Post-2007 drawdown: magnitude, duration, extremity

- HML value factor drawdown, 2007 to mid-2020: –55% (Arnott, Harvey, Kalesnik, Linnainmaa, *Financial Analysts Journal* 2021 — Graham and Dodd Scroll Award winner). - Duration: 13.5 years (2007–mid-2020), described by the authors as "the longest and deepest since 1963." - As of June 30, 2020, value's cheapness relative to growth reached the 100th percentile of historical valuation extremes.

What explains it: structural break, mismeasurement, or noise

Three explanations are debated in the sourced material, and they are not mutually exclusive:

Intangibles mismeasurement (Arnott et al. 2021). Book-to-price undercounts intangible assets (R&D, brands, patents) that make up a growing share of firm value. Capitalizing intangibles into the book-value measure: - shrinks the drawdown from –54.8% to –43.0% (about one-fifth less deep); - shrinks the duration from 13.5 years to 3.5 years (about three-quarters shorter); - the intangibles-adjusted value measure "outperforms the traditional measure by a wide margin," especially post-1990.

Valuation-spread explanation (same authors). A return decomposition finds that changes in the valuation spread between growth and value portfolios explain the *entire* drawdown — i.e., value didn't get structurally worse, growth got structurally more expensive relative to it, "with room to spare."

Statistical indeterminacy (Fama & French / Chicago Booth). As above: 28 years per sub-period is not enough power to rule out either explanation. The material found does not resolve whether the 100th-percentile cheapness by mid-2020 predicts mean reversion or whether growth's re-rating is now permanent — this is marked unverified in the notes; no source in hand settles it.

What does NOT work as an argument here

Citing only the original 1963–1991 Fama-French numbers (0.36%/0.58% monthly) as proof that value "works" is not supported by the material collected: those numbers describe a sample period that ended over three decades ago, and the post-1991 replication numbers (0.05%/0.33% monthly) are the ones that actually bear on whether the effect persists today. A page or claim that quotes only the pre-1991 figures without the post-1991 ones is citing an out-of-sample-untested number as if it were current evidence — exactly the pattern Out-of-Sample vs Post-Publication Decay: The Two Numbers That Tell You If a Premium Is Real warns against.

Transaction-cost survival of the post-1991 premium (0.05–0.33% monthly) is not addressed in the sourced material and is marked unverified in the notes — see Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one for the general method to test whether a premium this thin clears realistic costs before treating it as investable.

Why this page matters for the wiki's objective

HML is the reference case for the whole wiki's evidence bar: it has decades of peer-reviewed numbers on both sides (strong pre-1991, weak-to-indistinguishable-from-zero post-1991), a documented multi-decade drawdown with an exact magnitude and duration, and competing explanations that the original researchers themselves say the data cannot adjudicate. It is neither cleanly "verified" nor cleanly "refuted" — it is the clearest illustration of what an honestly *undetermined* factor looks like once the numbers are actually laid out, which is the standard What Counts as an Edge Here: The Evidence Bar This Wiki Applies to Every Technique asks every page to meet.

Related

- What Counts as an Edge Here: The Evidence Bar This Wiki Applies to Every Technique — HML is the worked example for why "large effect size" and "statistically resolved" are different claims; this page supplies the numbers that separate them. - Out-of-Sample vs Post-Publication Decay: The Two Numbers That Tell You If a Premium Is Real — the 86%/43% declines from the 1963–1991 baseline are exactly the kind of out-of-sample shrinkage that page's framework (McLean-Pontiff) is built to explain; the McLean-Pontiff decay figures (26% out-of-sample, 58% post-publication) are the same order of magnitude as HML's own decline. - Multiple Testing: Why t>1.96 Is Not Enough — the bar this wiki uses to grade a factor's significance — French's "there is no way to tell" verdict on a two-sub-period comparison is the direct motivation for demanding higher significance thresholds than t>1.96. - The Size Factor (Small-Minus-Big) — a textbook case of post-publication decay this wiki uses as a yardstick — a sibling Fama-French factor from the same era with its own, separately documented decay story; useful to compare whether value's post-1991 shrinkage is a factor-specific story or a pattern across the original three-factor set. - Transaction Cost Accounting — the arithmetic that separates a real edge from a paper one — needed to answer the still-unverified question of whether the shrunk 0.05–0.33% monthly post-1991 premium survives real trading costs at all.