The factor zoo, audited
Academics and quant shops have documented well over 400 signals that supposedly predict which stocks beat the market — value, momentum, quality, low-volatility, dozens of accounting ratios. This site rebuilds 18 of the most famous ones from free data (Ken French’s portfolios and AQR’s factor library), and checks each against the same blunt test: does it still work in the years after it was published?
The short answer, matching the academic literature (McLean & Pontiff, 2016): the average premium falls by roughly 69% post-publication, and once you account for how many things were tested, only a handful clear the bar.
Every anomaly, plotted by the year its paper came out and the t-statistic of our replication. Toggle to see how the picture changes when you only count returns earned after publication. Click any bubble for the full record.
Bubble size ∝ annualized long-short return. A t-stat above ~2 is the classic bar for “publishable”; above 3 is the bar once you account for how many things were tested.
How much of each premium survived. Bars past 100% mean the strategy actually lost money after its paper was published.
Drop from the in-sample premium to the post-publication premium. 100% = the premium completely disappeared. Click a bar for the full record.
Names that clear a t-stat of 3 on the full sample and survive a multiple-testing correction — plus the low-risk anomalies, judged on their CAPM alpha.
Momentum · published 1999 · +6.4%/yr, 31% decay
Low risk · published 2014 · +7.8%/yr, 49% decay
Reversal · published 1990 · +9.5%/yr, 100% decay
Momentum · published 1993 · +7.2%/yr, 56% decay
Quality · published 2013 · +3.9%/yr, 23% decay
Low risk · published 1972 · +5.4%/yr risk-adjusted alpha
Take a premium and t-stat from a paper, an ETF fact sheet, or a newsletter, and run them through what this site actually measured: the average post-publication shrinkage, and the significance bar once you account for how many signals the field has tested.
The average published premium here fell 69% from its in-sample value to what it earned after publication. Applying that same shrinkage to 8.0%/yr leaves about 2.5% — before your own commissions, spread, borrow cost, and taxes, which have historically taken most of what’s left of a long-short equity strategy.
Enter a t-statistic to check it against the bar for multiple testing. For a data-mined predictor, Harvey, Liu & Zhu (2016) put the threshold near 3.0, not the usual 2.0.
Reference point: of the 18 well-known anomalies replicated on this site, only 5 still carry a full-sample t-stat above 3. This is a rough screen from aggregate history, not a verdict on any specific strategy — a genuinely new signal, or one with a real risk-based story, can still be worth holding.
Sortable. The columns that matter: full-sample t-stat, the in-sample vs post-publication premium, the decay between them, and whether a risk-adjusted alpha survives the other known factors.
| Anomaly | Pub.↑ | Full t | In-sample | After pub. | Decay | α vs factors | Verdict |
|---|---|---|---|---|---|---|---|
Low beta / betting-against-beta Low risk · 1926–2012 | 1972 | 2.2α | +4.0% | +6.2% | -57% | −0.9 | Faded |
Value (earnings-to-price)△ Value · 1957–1971 | 1977 | 2.4 | +4.0% | +2.0% | 51% | −2.0 | Faded |
Dividend yield△ Value · 1936–1977 | 1979 | 0.6 | +1.9% | +0.5% | 74% | 0.1 | Gone |
Size (small-minus-big)△ Size · 1936–1975 | 1981 | 1.6 | +3.1% | +0.4% | 86% | 1.7★ | Gone |
Value (book-to-market) Value · 1963–1990 | 1985 | 2.3 | +5.1% | +2.1% | 60% | 0.7★ | Faded |
Long-term reversal (2–5 years)△ Reversal · 1926–1982 | 1985 | 2.0 | +10.0% | +0.5% | 95% | −0.3 | Gone |
Short-term reversal (1 month)△ Reversal · 1934–1987 | 1990 | 4.2 | +14.2% | +0.0% | 100% | 1.7 | Holds up |
Momentum (12–2) Momentum · 1965–1989 | 1993 | 3.9 | +9.8% | +4.3% | 56% | 4.3★ | Holds up |
Value (cash-flow-to-price)△ Value · 1963–1990 | 1994 | 2.1 | +6.4% | -0.1% | 101% | −2.1 | Gone |
Accruals△ Quality · 1962–1991 | 1996 | 2.8 | +4.5% | +1.9% | 57% | 2.5 | Faded |
Industry momentum△ Momentum · 1963–1995 | 1999 | 6.1 | +7.8% | +5.4% | 31% | 5.7 | Holds up |
Idiosyncratic volatility△ Low risk · 1963–2000 | 2006 | 1.7 | +9.0% | +0.6% | 94% | 1.8 | Gone |
Total volatility△ Low risk · 1986–2006 | 2007 | 1.1 | +7.0% | -1.1% | 116% | 1.0 | Gone |
Investment / asset growth△ Investment · 1968–2003 | 2008 | 2.9 | +5.5% | +0.2% | 96% | 2.9★ | Gone |
Net share issuance△ Quality · 1970–2003 | 2008 | 2.6 | +6.5% | +3.4% | 48% | 2.2 | Faded |
Profitability (gross / operating) Profitability · 1963–2010 | 2013 | 2.8 | +3.3% | +2.2% | 33% | 4.1★ | Faded |
Quality minus junk Quality · 1957–2016 | 2013 | 3.6 | +4.3% | +3.3% | 23% | 5.3 | Holds up |
Betting-against-beta (AQR factor) Low risk · 1926–2012 | 2014 | 5.7 | +8.1% | +4.2% | 49% | 2.2 | Holds up |
“α vs factors” = t-stat of the risk-adjusted alpha against the Fama–French 5 factors + momentum (with the anomaly’s own factor removed where it is one, ★). Newey–West standard errors throughout. △ = our value-weighted / univariate construction is milder than the original paper’s — hover for detail. In-sample and post-publication columns are our replication over each window; “decay” is the drop between them.
Data as of 2026-10-16. Monthly long-short (decile-spread or published-factor) returns, value-weighted, US equities, gross of transaction and shorting costs.