Indicator Audit · Batch 3

The strongest published evidence, judged on its own claims

We audited six strategies with some of the strongest support in the literature: volatility-managed equity, the low-volatility anomaly, time-series momentum, Faber’s timing model and risk parity. Each was judged on the objective its own source claims, declared and sealed before the data were fetched. None of the six meets its claim under our test (0 of 6 survive). One came close, and only half of its claim held.

The closest call: Faber timing on EEM, VNQ and TLT

These are markets we had never examined. The timing rule cut drawdowns, robustly. The maximum drawdown was 14.5% against 43.9% for holding the basket, the bootstrap 5th percentile of the reduction was 3.5%, and the Holm-adjusted p is 0.0006. Every check passed except one.

Faber’s claim is lower drawdowns at similar returns. Before the run we declared that giving up at most 1.0% a year would count. It gave up 3.2% a year. So its grade is capped at C, and it does not count as meeting the claim.

The rest

Volatility-managed SPY (Moreira & Muir) raised the Sharpe ratio from 0.65 to 0.79. But the improvement did not survive the bootstrap and the 12-variant correction (grade D).

Low volatility (a sector rule and USMV), time-series momentum and risk parity did not improve the Sharpe ratio on this history. Time-series momentum compounded faster than holding the same ETFs only through leverage financed at 0%.

We added a new objective, “improve Sharpe vs benchmark”, as rubric v1.2 and sealed it before this batch. No earlier audit was re-graded. Statistics on past data only; not advice, and not a verdict on the papers.

Equities and equity ETFs

RuleMarketObjective (its own claim)VariantsGradePrimary resultHolm pSurvives
Volatility-managed equity (Moreira & Muir)SPY(c) improve Sharpe12DSharpe 0.79 vs 0.65 (improvement +0.14, 5th pct −0.01); holdout +0.17 → +0.071.0000no
Low-volatility sectors (low-volatility anomaly)9 sector SPDRs(c) improve Sharpe9FSharpe 0.62 vs 0.51 (improvement +0.12, 5th pct −0.06); holdout +0.20 → −0.081.0000no
Minimum-volatility ETFs (USMV, SPLV)USMV / SPLV(c) improve Sharpe2FSharpe 0.88 vs 0.92 (improvement −0.04, 5th pct −0.25); holdout +0.02 → −0.231.0000no

Multi-asset and bonds

RuleMarketObjective (its own claim)VariantsGradePrimary resultHolm pSurvives
Time-series momentum (Moskowitz, Ooi & Pedersen)SPY, EFA, EEM, IEF, TLT(c) improve Sharpe8FSharpe 0.53 vs 0.65 (improvement −0.13, 5th pct −0.52); holdout −0.02 → −0.281.0000no
Faber GTAA timing on unseen markets (EEM, VNQ, TLT)EEM, VNQ, TLT(b) reduce drawdown8Cmax drawdown 14.5% vs 43.9% (5th pct of the cut 3.5%); return cost 3.2%/yr vs declared 1.0%0.0006no
Risk parity stocks/bonds (Asness, Frazzini & Pedersen)SPY / TLT(c) improve Sharpe8DSharpe 0.79 vs 0.79 (improvement +0.00, 5th pct −0.17); holdout +0.17 → −0.161.0000no

Sources. Moreira & Muir (2017), J. Finance; Blitz & van Vliet (2007), JPM, and Baker, Bradley & Wurgler (2011), FAJ; Moskowitz, Ooi & Pedersen (2012), JFE; Faber (2007), J. Wealth Management; Asness, Frazzini & Pedersen (2012), FAJ. Full citations are in each report.

Seals. Rubric v1.2 was sealed at 20:33:45 UTC; the preregistration and six declarations at 20:39 UTC on 2026-09-25, before the new data were fetched. Preregistration SHA-256 2142d17c29db1f2b0080e225cfbcd15ba67582b4122adcc6c78b0aad1325d690; seal record 90ff232e75d552219386a15d1bc0990a86b40a925d2a7dc1405e7be67995e009.

Disclosures. Before this batch we had seen prices for SPY, EFA, IEF and the sector SPDRs, but not for EEM, TLT, VNQ, USMV or SPLV. The volatility-managed and sector audits therefore used prices we had seen. Cash earns 0% and borrowing costs 0%. Commodities are excluded from time-series momentum. There were no deviations from the preregistered protocol.

Prices from Yahoo Finance’s public chart endpoint; we publish derived statistics only, never raw prices; no redistribution.

These audits analyze published rules on historical data for illustration. They are not recommendations to trade any of these strategies, and a grade is not a prediction.