At record prices, the decade ahead offers a much worse reward for its risk, and that affects almost every stock investor's whole portfolio. But prices can't tell you when a decline starts. So the useful question is how to hold your stocks through the next decline without predicting anything. This report tests one answer, the practical solution, over 100 years of US data, plus Canada and Japan. The evidence on valuations themselves is in the main report: Why Berkshire (sometimes) holds more cash than listed stocks.
Signals are known at month-end and applied to the next month. CAPE is lagged one extra month for publication delay, and every switch costs 0.1%. The full-exit version is identical except that step 4 moves all of the stock allocation to cash. The exit was changed on 2026-09-29, after two Red Team reviews: the first version re-checked valuation monthly and went back to full stocks in the middle of the 1930 and 2008 crashes.
Halving is easier to live with. Big switches are hard for investors and advisors to adopt and to hold. The same signals with a smaller response: when the rule fires, cut stocks by 50% (the practical solution) or 100% (the full-exit version). Each row is compared with its own buy-and-hold. The 60/40 rows put the freed money into bonds. Rebalanced monthly; 0.1% per switch.
Halving is the practical sweet spot. For an all-stock investor the practical solution earned vs a year and cut the worst decline from to . For a 60/40 investor the practical solution matched buy-and-hold ( vs ) with a worst decline of vs . Sheltered accounts only gives the same result: running the full-exit version on the half of a portfolio held in an RRSP or TFSA is, before tax, the practical solution, and the taxable half never realizes a gain.
Why this is measured first. There is no 10-year result for an investor who abandons the plan in year 3. The Stick-to-Plan Factor counts the share of all 2-year periods spent in a comfort zone: never more than 20% below the last high, and not behind where the investor was 2 years earlier. It is a descriptive comfort proxy, not a measured adherence probability; results for 1-, 3- and 5-year windows are in the project files.
Monthly rules can't catch a one-month crash. Total return from the pre-crash peak month to the trough month. The 1987 crash took one month, too fast for any monthly rule. In 1929–32 the first version of the exit switched off as prices fell and re-entered early; the adopted exit stays out until the trend recovers.
The trend rule was popularised by Faber (2007) and the valuation-plus-momentum idea by Asness, Ilmanen and Maloney (2017). Results after publication are the honest out-of-sample test.
After publication, a mostly rising market. Each cell: return per year · worst decline. After publication, both rules lagged buy-and-hold on return. The trend rule still cut the worst decline sharply (2007–2026: vs ). The full-exit version earned vs a year over 2007–2026.
Every combination of valuation threshold and moving-average length, plus a rate-adjusted valuation signal. A result that holds across neighbouring settings is less likely to be a fluke of one choice.
Neighbouring settings agree. All valuation + trend settings of the full-exit version returned to a year vs buy-and-hold, with Sortino between and (buy-and-hold: ). Longer averages switched less and did slightly better. Trend-only settings all gave up about points a year for roughly half the worst decline.
Taxes erase the trend rule's edge in Canada. Signal: US CAPE (there is no complete Canadian series), trend on the TSX itself. In a taxable account the trend rule realizes gains at every exit, and its edge disappears. The full-exit version, with its valuation filter, switches far less, and roughly matched buy-and-hold after tax.
Japan: out early, and much of the fall avoided. The trend rule exited in , from the peak, and roughly halved the worst decline over 76 years at about the same return. Japan's valuation filter (dividend yield below 1%) was armed from (the first month the yearly yield figure was available), about years before the top.