Valuation + Trend Rule

Objective — cut the damage from a major decline that starts at extreme valuations, with a written rule set in advance. It replaces a judgement made under greed or fear.

Risk level — 1. The CEO set level 1 on 2026-09-29 (Market-timing-Red-Team round 4), replacing the level 0 agreed on 2026-09-25. Formal Risks sign-off is still pending: the charter routes tier assignment through Risks first, so this stays a draft until Risks confirms. Risk justification — level 0 was rejected by both Red Team reviews (Claude pass 1 and OpenAI GPT-6 Sol pass 2, 2026-09-28/29). The rule does not only reduce exposure; it also restores it. As written, it can restore full stocks in the middle of a continuing crash (Dec 1930, Feb 2008). And even a reduction carries client risk: opportunity loss (the worst 10-year period cost $69,020 per $100,000, for the 2016-05 start), tax realization, and a harmful override or abandonment. It is not level 3 on the timing-judgement test, because it needs no judgement: it is mechanical, month-end and pre-committed. Benefit justification — unverified — needs Talbot. Jurisdiction — Canada for the account guidance (use RRSP/TFSA). The evidence is US 1926–2026, the S&P/TSX 1957–2026 and Japan 1950–2026. U.S. applicability of the account guidance is unverified. Name — Valuation + Trend Rule is a working label, not a coined name. It needs the CEO (JOB_DESCRIPTION scope 3).

Prerequisites

• A written allocation (e.g. 100/0 or 60/40) and a written copy of the rule, signed before it is needed.
• Ideally, the switching happens inside registered accounts (RRSP, TFSA), where exits realize no tax.

Mechanism

Once a month, at month-end:

1. Valuation filter: is the Shiller CAPE at or above the 90th percentile of its own history to date?
2. Trend test: is the index’s total-return level below its 10-month simple moving average?
3. If both: cut the stock allocation, either fully to cash (T-bills, a high-interest savings account or a cashable GIC) or, in the halving variant, by half.
4. Exit (latched; CEO decision 2026-09-29, after both Red Team reviews): once cut, stay cut until the total-return level closes back above its 10-month average, whatever valuations do. Valuation only starts a cut; the trend alone ends it.
5. Otherwise: hold the normal allocation.

Benefits

• Smaller worst cases. US 1926–2026, 100/0, halving with the adopted latched exit:
  • worst decline −62% vs −84% for buy-and-hold;
  • 60/40 (30/70 when it fires): −40% vs −63%;
  • robust to signal delay: −62% whether CAPE is known 1, 3, 6 or 12 months late;
  • from Dec 1999, $100,000 became $144,121 (worst decline −30%) vs $96,264 (−50%) doing nothing;
  • in 2000–02 the full-exit version lost −8% vs −41%.
• Return effect: not established. Halving (latched) earned 10.97% vs 10.33% a year over 1926–2026. But:
  • the block-bootstrap 90% interval for the gap is −0.30% to +1.70% a year, which includes zero;
  • without 1929–32 the gap is −0.15% a year;
  • since 2017 it lagged: 12.4% vs 14.9%.
  • Present it as protection, not a return engine.
• Behavioural. A pre-committed exit and re-entry replaces the two common mistakes, buying heavily late in a bull market and selling after a crash.

Risks

• Crashes that begin from ordinary valuations are not insured. The filter wasn’t armed in 1973–74 or 1987. In 2008–09 it was armed at the Dec 2007 trend break, then disarmed from Feb 2008 to Nov 2009 as valuations fell, putting the holder back in for most of the crash (−48% vs −50% from Dec 1999). This is the exit flaw below.
• Trailing, often, is the price of the protection. Halving (latched), for a new adopter at each of 1,081 ten-year start months 1926–2016, 100/0:
  • 316 ahead, 373 identical (it never fired), 392 behind; median $0.
  • Worst period: −$79,352 per $100,000 (1990-09 start).
  • By start era: 1926–59: 123 / 180 / 99; 1960–89: 1 / 193 / 166; 1990–2016: 192 / 0 / 127.
  • From top-5% valuation starts: 106 ahead, 28 behind.
  • The superseded exit was ahead more often (512 ahead / 196 behind), but its worst decline was −80%.
  • An investor must expect to trail for years and keep following the rule.
• Today the filter is almost always armed (97% of months this decade), so in practice the rule currently behaves like a plain trend rule, with that rule’s higher switching. CAPE’s level has drifted up (median about 14–15 before 1990, 24.8 for 1990–2016; cape-timing L12), so a since-1881 threshold may stay armed for years.

Failure modes

• Whipsaws: 22 exits in 100 years reversed within three months, several of them missing rallies of more than 5%.
• Very fast crashes (Oct 1987, Mar 2020) happen inside a single month, too fast for a monthly rule.
• Abandonment: the most common real-world failure is giving up the rule after years of trailing.

Costs

• Forgone return in rising markets (see Risks).
• Trading costs (modelled at 0.1% per switch).
• In a taxable account, each exit realizes gains. TSX 1957–2026, $200k Ontario bracket: the plain trend rule fell from 9.8% to 7.3% a year after tax, below buy-and-hold’s 7.9%.

Tax considerations

Use it inside an RRSP or TFSA, where switching is tax-free. Applying the all-or-nothing rule to only the registered half of a portfolio is, before tax, the same as the halving rule, and the taxable half never realizes a gain. Interest on cash (T-bills, HISA, GICs) is fully taxable outside registered accounts.

Who it may suit

• Financially independent investors, and those near or in retirement, for whom the worst case matters more than the last point of return.
• Investors who know they would sell in a crash and want a rule that decides for them.

Who should avoid it

• Long-horizon accumulators who can genuinely sit through −50% or worse. Buy-and-hold remains very hard to beat on expected wealth.
• Anyone unlikely to follow the re-entry signal. A rule followed on the way out and not on the way back in is worse than no rule.
• Taxable-account-only investors, unless they use the halving variant and accept the tax drag.

Implementation outline

1. Choose the base allocation and the response: halve, or go to cash.
2. Choose the accounts: registered first.
3. Put the monthly check on the calendar (month-end total-return level vs its 10-month average; CAPE percentile).
4. Write down the rule and the re-entry condition, and sign it.
5. Review the rule annually, not after a scary month.

Evidence status

documented — backtested in ~/projects/cape-timing (cape-timing): US CRSP 1926–2026 (monthly, no look-ahead, 0.1% per switch), a robustness grid (CAPE threshold 80–95th × moving average 6–12 months: all 16 settings earned 10.1–11.2% a year vs 10.3%), the S&P/TSX 1957–2026 and the Nikkei 225 1950–2026. Reports: Market Timing at High Market Valuations (https://claude.ai/artifact/BfA8jRojdFHtLArKBj3U8w) and Valuation + Trend Rule (https://claude.ai/artifact/4xd4oHHcwSEwCiJ6PqhrKB). Not modelled in sd-math.

Counterarguments

• “It’s market timing.” Yes: rules-based, pre-committed and modest. Asness, Ilmanen and Maloney (2017) conclude that valuation timing alone doesn’t pay, but that adding momentum helps. “Sin a little.”
• “Six rules and 23 settings were tested.” Multiple-testing risk is real. The defence is the plateau across settings and the prior literature (Faber 2007; Asness 2017; Clare et al. 2017), not the best cell.
• “The edge is mostly 1929–32.” Largely true for the full-switch version’s return edge. The halving variant’s case rests on drawdown reduction over 1926–2026, not on extra return; era-by-era results for halving are not yet computed.

Variants

• Halving (the recommended practical form): cut stocks by 50% when the rule fires; 100/0 becomes 50/50, and 60/40 becomes 30/70.
• Trend only (no valuation filter): larger drawdown reduction (the US worst decline fell from −84% to −43%) at about 0.7 points a year of return.
• Filter exit (superseded 2026-09-29): cut only while both conditions hold. It was ahead more often, but its worst decline was −80%, because it went back to full stocks mid-crash.
• Rate-adjusted valuation (Excess CAPE Yield ≤ 10th percentile): armed in only 3% of months 1926–2026, and not armed today.
• Leverage side: deleverage when the rule fires, and re-lever when it clears. See pre-decline-deleveraging.

Related strategies

pre-decline-deleveraging · post-decline-deployment · buy-more-low · conservative-leverage-ratio

Sources

• cape-timing — code, data provenance and gates
• Asness, Ilmanen & Maloney (2017), Market Timing: Sin a Little, JOIM
• Faber (2007), A Quantitative Approach to Tactical Asset Allocation (10-month SMA rule), as cited
• Clare, Seaton, Smith & Thomas (2017), Reducing sequence risk using trend following and the CAPE ratio, FAJ
• Zakamulin (2014, 2016) — look-ahead and cost biases in moving-average backtests
• Kitces & Pfau (2015), Retirement Risk, Rising Equity Glide Paths, and Valuation-Based Asset Allocation, JFP

Open questions

• The coined name (CEO).
• Tier confirmation: the CEO set level 1 on 2026-09-29 (it had been 0); Risks sign-off is pending.
• Suitability exclusions (OpenAI Red Team, 2026-09-29, to be confirmed): near-term withdrawals or a binding spending floor; large embedded taxable gains without a tax-lot plan; too little registered-account room; a portfolio unlike the tested stock/cash mix; unwillingness to follow the rule through years of underperformance.
• After-tax, household-level test of the halving version: to be done in the lab task (sdc-market-timing-lab). Until then, taxable accounts are untested.
• Exit rule Decided 2026-09-29: latched (CEO, after both Red Team reviews; see Market-timing-Red-Team). The first version re-checked valuation every month and disarmed mid-crash (Dec 1930; Feb 2008 to Nov 2009). Code: cape_timing.vt.fired(latched=True).
• A Canadian valuation signal: the backtest uses US CAPE for the TSX, because no complete Canadian CAPE series exists.
• Client-facing use needs supervisor pre-approval (MDW-review §6.3). See the MDW offering successor task.