Validation

Factor library validation record, v0.1.

The publication rule, the result of applying it to library v0.1 on data through 2026-08-31, and the cross-validation against public benchmarks. The full record is a public PDF.

Download the validation record (PDF, 11 pp)   Methodology changelog

Publication rule

In force: v0.1.

A factor is eligible for publication as a return factor only if (i) the Fama-MacBeth t-statistic on its risk premium exceeds 2.0, (ii) its information ratio in the holdout period is positive, and (iii) its decile spread remains positive net of 20 basis points per side of transaction cost. A factor that fails is still computed and reported as a risk descriptor.

Status vocabulary. PASS: meets all criteria of the publication rule in force and is eligible for publication as a return factor; MONITOR: does not meet the rule but shows either a Fama-MacBeth t-statistic above 1.0 or a positive holdout information ratio; computed and reported as a risk descriptor, not presented as a source of return; FAIL: does not meet the rule on any count; computed and reported as a risk descriptor pending point-in-time data.

Proposed, not in force. Proposed for library v0.2: all t-statistics Newey-West adjusted (lag 6); adds (iv) holdout-period joint Fama-MacBeth t-statistic > 1.0. Effective only when approved and recorded in the changelog.

Results

1 of 4 families meet the rule.

FamilyImplemented definition (library v0.1)FM t-statIR (full)IR (holdout)Turnover / moNet @20 bpsStatus
MomentumTotal return from month-end t-12 to month-end t-1 (skips the most recent month)0.50-0.020.6155%-3.0%MONITOR
Low volatilityNegative of annualized standard deviation of daily returns over trailing 252 trading days-3.11-1.01-1.1917%-27.6%FAIL
Size (small minus big)Negative log market capitalization (current shares x month-end price; demo approximation)3.031.270.4411%+15.4%PASS
Short-horizon reversalNegative of trailing 21-trading-day total return-0.12-0.09-0.79172%-9.5%FAIL

Construction sample 2016–2022, holdout 2023-01-01 onward; 127 monthly cross-sections; 482 stocks. Fama-MacBeth t-statistics are uncorrected for autocorrelation. Net @20 bps is the annualized long-short return after a linear cost haircut of 20 basis points per side on measured turnover.

Size caveat. Size is the only family that meets the publication rule in this build. It is also the family whose demo input is not point-in-time — market capitalization is approximated as the current share count multiplied by the adjusted price — so it should be treated as the least reliable of the four until as-reported share counts are licensed.
Cross-validation

Against the Kenneth R. French Data Library.

In-house spreadBenchmarkMonthsPeriodCorrelationBeta to benchmarkSign agreement
MomentumFrench UMD1262016-02 to 2026-070.9001.3387%
Size (small minus big)French SMB1262016-02 to 2026-070.5980.8067%
Momentum, sector-neutral variantFrench UMD1262016-02 to 2026-070.841

The benchmark series are value-weighted 2×3 sorts on the full CRSP universe; the in-house spreads are equal-weighted deciles within the demo universe, so correlation rather than level agreement is the relevant test. The size spread correlates at 0.60 with SMB; the lower figure is expected because SMB spans the full market while the demo universe contains large caps only, so the in-house spread measures size dispersion within the S&P 500 rather than small-versus-large across the market. Sector neutralization removes the component of momentum that is a bet on sector rotation; both variants are carried with their own validation.

How to reproduce the comparison →