Arguments About Personality - a research note on the Fama-French series

Four of the most celebrated arguments in quantitative finance, run through one honest machine: the same universe, the same yardstick, twenty sealed years each, and - three times out of four - every dividend, every trading cost, and a benchmark allowed to keep its own dividends. Not one verdict was rankable. That is not the disappointment it sounds like.

Over one week we ran four famous cross-sectional signals through the same machine. Each became two portfolios of thirty S&P 500 stocks, selected point-in-time, equal weighted, re-selected annually, walked across twenty sealed one-year windows from 2006 through 2025 - each window registered before it ran, each result frozen when it landed. One book appeared in every contest: thirty stocks ranked on earnings yield. Plain cheapness, the oldest screen in the book. It was the yardstick.

The first study put the yardstick itself on trial. Academic value was built on book-to-market: Fama and French wrote in 1992 that it absorbs the earnings yield, and forty years of factor models took them at their word. Twenty sealed years of large caps found the two measures selecting almost disjoint portfolios - by 2023 they shared exactly one stock in thirty - diverging by ten to twenty points in single years, and finishing a statistical dead heat. The absorption claim, where most invested money actually sits, is moot.

Then came the challengers, each measured in total returns with real costs.

Dividend yield - the income aisle's favorite - collected 4.6 percent a year in cash, more than double the index's rate, and compounded almost exactly at the honest index's pace, on a worse Sharpe than the index. The cash was real and spendable; the price leg gave it back. The gap to the cheapness book is not statistically certifiable - the two books diverge by about twelve points in a typical year, and at that noise roughly two hundred and eighty years would be needed to certify the observed lean - but the structure is not noise: the market handed the income out of one pocket and took it from the other.

Gross profitability - Novy-Marx's 'other side of value', Fama-French's RMW - produced the series' best risk-adjusted record: the highest Sharpe of the quartet on volatility below the index's own, the only book of the three total-return studies to beat the honest benchmark on return and risk at once (a description of one path, not a certified premium). Against the cheapness yardstick: ten windows each, a coin flip. The two books shared a median of ONE name in thirty - the most disjoint pair in the series - and the profitability book won every crisis on its record, COVID by thirty-eight points.

Net share issuance - the buyback signal Fama-French ranked among the strongest anomalies - finished the tightest race of all: two tenths of a point of annual return apart after twenty years. Its holder was paid an index-like two percent in dividends and received the rest through a quietly shrinking float - the mechanical opposite of the dividend book. Its one famous vice is printed in its ledger: in the 2007 window, corporate America repurchasing record volumes at peak prices into the coming crash, it trailed cheapness by fourteen points.

Four contests. Four statistical ties. Every lean against the yardstick - 0.2 to 1.4 points a year - inside the noise. If the series had been run to crown a winner, it failed four times. It was not run to crown a winner, and the ties are not the finding.

The finding is what stayed measurable after the ranks dissolved. These portfolios are not variations on a theme - they are nearly strangers. The profitability and cheapness books agreed on one stock in thirty; the dividend and cheapness books on four; in several windows, on none at all. They pay in different currency: 4.6 percent cash from the dividend book, 1.3 from the profitability book, a shrinking share count from the buyback book. They shelter in different storms: profitability won every crisis on its record, while the dividend book cushioned the rate shocks and failed exactly once - in 2008, when high yield meant bank yield about to be cut. Yield is a promise written by the past; earnings are the capacity to keep it.

One pattern crossed the studies uninvited: since 2022, plain cheapness has won every single window against both quality and buybacks - nine consecutive contests across two studies. (Honesty requires the footnote: two of that era's windows had been run as unregistered engine rehearsals before one of those studies sealed; no design changed, and the sweep's remaining windows ran sealed-first. It is an observation found in the record, not a claim made before it.) Whatever regime arrived with higher rates has, so far, paid the cigar butts.

So the quartet's answer to 'which factor is best?' is that the question is malformed at this altitude. Among the five hundred most-watched stocks in the world, measured in the units a holder actually receives, the famous signals are indistinguishable in destination and utterly distinguishable in character. The choice between them is not a forecast of return. It is a choice of experience: what you hold, how you are paid, when you suffer, and whether your risk can be drawn in advance. Twenty sealed years cannot rank the returns. They can describe the personalities with precision.

The cross-section's famous arguments, measured honestly, are arguments about personality.

How to read this note

Every study cited here is a two-arm comparative walk on the point-in-time S&P 500: thirty names per arm, equal weight, long only, annual re-selection, twenty one-year windows anchored each January from 2006, each window's hypothesis registered before running and its report frozen at execution. The dividend, profitability and issuance studies are measured in total returns (ex-date dividend credits, split-adjusted amounts) net of ten basis points per one-way traded dollar, against an equal-weight total-return benchmark; the book-to-market study is price returns against a price benchmark, labeled as such, and its row in the table above must be read on that basis.

'Bootstrap' in the figures and table is the fraction of 2,000 block-resampled twenty-year paths on which the first arm finishes ahead - a path-level statistic, not a daily win rate. Every 'cannot certify' in this note is that bootstrap speaking. This note registers nothing of its own; the four full records - every window, every holding, every projection cone - are open at the links above.