One rule, published in 2007 and copied everywhere: stay invested when the month-end price is above its 10-month average, otherwise switch to Treasury bills. Faber applied it to five asset classes at equal weights.
Equity-like returns with much smaller drawdowns and volatility. Not a higher return.
| Market | Unseen period | Verdict | Rule / buy-and-hold | Max drawdown rule / buy-and-hold | Proof |
|---|---|---|---|---|---|
| Developed ex-US stocks | 2007-07 → 2026-08 (19.1 y) | No proof of an edge yet | +106.7% / +170.3% | 26.3% / 55.9% | SAP-4497a1a6480b |
| US stocks (total market) | 2007-07 → 2026-08 (19.1 y) | No proof of an edge yet | +429.4% / +665.1% | 18.8% / 50.3% | SAP-4f3589dd7254 |
What the theory brings, measured. Over 19.1 unseen years (since July 2007, after the paper came out), the 10-month average earned less than buy-and-hold (+429.4% vs +665.1% on US stocks), but its worst drawdown was 18.8% while the market lost 50.3%: divided by 2.7. Same reading outside the US (26.3% vs 55.9%). That is exactly Faber's promise: much shallower drawdowns, paid for with part of the upside; the Lab finds no gain above buy-and-hold.
Rule written and preregistered on 12/10/2026, before any proof, with its source's parameters; one trial per market, nothing tuned; judged in certified mode by Lab 270, which cuts the unseen period itself: everything after the source's publication. "Rule / buy-and-hold": gain of the rule and of buy-and-hold on the same bars, 10 bp fee per order. Proofs SAP-4497a1a6480b, SAP-4f3589dd7254. Gaps with the source: One asset class per proof (not the five-asset portfolio); out of the market, cash at 0% instead of bills (unfavourable to the rule); gold and Shiller's S&P are left out because their monthly values are averages of the month. Data: Ken French Data Library, FRED (Board of Governors). Research on past data, not advice; no future return is promised.
Faber, M. T. (2007), "A Quantitative Approach to Tactical Asset Allocation", Journal of Wealth Management 9(4), 69-79, DOI 10.3905/jwm.2007.674809; update: JPM 44(2), 2018, DOI 10.3905/jpm.2018.44.2.156.