What works
What to do instead — the levers that actually work
Since prediction doesn't beat the index, these are the only things shown to help: front-load the NISA quota, don't panic-sell, minimise cost and tax.
FINDING01What's the only thing that survived this research program?Two survivors out of 15+ strategies — neither raises returns.PROVEN
Trend overlay (crash insurance)
The two results that survived the research program — both lower drawdown, neither raises return. Current regime, computed from price.
12-month trend — broad-equity core
Equity + gold diversifier
Suggested allocation
- S&P 500 50%
- Gold 50%
Track record (backtest, after costs)
| Strategy | CAGR | max DD | Sharpe |
|---|---|---|---|
| S&P 500 buy & hold | +11.3% | -50.8% | 0.81 |
| Equity + gold trend | +10.6% | -13.5% | 1.08 |
| 60 / 40 | +8.7% | -28.5% | 0.90 |
Diversifier correlation to equity: 0.56 · 2002-07-31 – 2026-06-30
Max drawdown in crises (equity → diversifier)
- 2008 GFC-50.8% → -13.5%
- 2022-23.9% → -9.2%
What the diversifier buys: a far shallower drawdown for a little less end-wealth — equity buy-and-hold still ends richest. Monthly, switch costs included.
Once a month: if the index's trailing 12-month return ≥ 0 stay invested, else hold cash until it turns positive. Cuts sustained-bear drawdowns roughly in half at ~zero long-run return cost; it does NOT raise return and does NOT help flash crashes. Only the S&P 500 overlay is historically validated (98yr); the other indices apply the same rule. Caveat (2026-06-05): validated for a taxable / generic account with instant re-entry — it does NOT survive the 新NISA wrapper, where de-risking forces a sale and re-entry is throttled by the ¥3.6M/yr acquisition cap; routing new contributions by trend also loses to plain DCA. For a NISA holder the answer stays plain buy-and-hold. Decision-support, not advice.
FINDING02Inside the one crash hedge that workedThe rule is one line. What it costs you is published too.USABLE (RISK-ONLY)
The one thing that worked: a 12-month-trend crash hedge
The single genuinely usable result in the whole program — risk reduction, not extra return. S&P 500, 1927–2026, monthly: when the trailing 12-month return turns negative, move to cash; re-enter when it turns positive.
It cut the worst drawdown of the last century from −86% to −52% at roughly zero long-run return cost (CAGR 6.1% → 6.0%), beating buy-and-hold in 4 of 4 sustained bear markets — but it does NOT help flash crashes (2020), and it does not survive the NISA wrapper.
| Sustained bear | Buy & hold maxDD | Trend hedge maxDD |
|---|---|---|
| Great Depression 1929–32 | −86% | −34% |
| Oil shock 1973–74 | −46% | −13% |
| Dot-com 2000–02 | −46% | −14% |
| Global financial crisis 2007–09 | −53% | −11% |
| COVID 2020 (flash crash) | −20% | −20% ✗ |
| 2022 de-rating | −25% | −13% |
| Full period 1927–2026 | −86% | −52% |
This is the honest high-water mark: a rule that halves-to-quarters the pain of slow, grinding bear markets for free. But it whipsaws on V-shaped crashes like COVID (sold low, missed the snap-back), it's a risk reducer not a return booster, and inside NISA the forced sale plus the ¥3.6M/yr re-entry cap break it. Real for a taxable account; for a NISA holder the answer stays plain buy-and-hold.
S&P 500 price index 1927–2026; ~1 switch/yr; dividends and cash yield omitted (opposing biases). USD; FX out of scope. Risk management, not a return or edge claim. Not advice.
FINDING03How much shallower do crashes get? (S&P 500, 98 years)The drawdown gap vs buy-and-hold, drawn out.CHART
A century of drawdowns: buy & hold vs the trend hedge
Monthly drawdown (% below the running peak), S&P 500 1927–2026. Buy-and-hold falls to −86% in 1932 and −53% in 2008; the 12-month-trend hedge stays far shallower in sustained bears (worst −52%) — at roughly zero long-run return cost. Risk reduction, not a return claim. Not advice.
FINDING04And with an equity + gold diversifier?Max drawdown, crisis by crisis, side by side.CHART
The 2nd usable result: equity + gold trend diversifier
Monthly drawdown, 2003–2026. The equity+gold 12-month-trend diversifier (worst −13.5%) sidesteps the crashes that take equity to −51% (2008) and −34% (2020) — higher Sharpe (1.08 vs 0.81), low correlation (0.55). Risk-adjusted, not extra return (it ends slightly behind equity), and it's more complex/higher-friction than a single fund. Not advice.
FINDING05So how should monthly contributions actually be set up?Removing decisions is the real edge — the autopilot blueprint.PLAYBOOK
Contribution autopilot — fill the NISA quota, front-loaded
No market timing. The only contribution edge is front-loading: deploy each year's quota as early as cash allows.
—
Why front-load
Across 1,159 rolling S&P 500 years, lump-sum / front-load beat monthly DCA 67% of the time (median +3.6%). Waiting for a dip LOST — cash drag, and the dip often never comes.
Funding from salary? Plain monthly DCA is fine — the edge only exists if you have idle cash to deploy. Never hold cash waiting for a dip.
Decision-support, not advice. Your numbers stay in your browser (localStorage) — nothing is sent anywhere.
FINDING06How often has front-loading won, exactly?A century of outcomes in one histogram — including the losses.100-YR DATA
Front-load vs DCA: how often does deploying early win?
Across 1,159 rolling 1-year S&P 500 cohorts, front-loading the annual quota beat spreading it monthly (DCA) in 67% of years (median +3.6%). Green = front-load wins; red = the down years where DCA wins. Markets rise ~2/3 of the time, so time-in-market beats waiting — and buy-the-dip (holding cash) loses to both. Decision-support, not advice.
FINDING07How do you guard against the biggest risk — yourself?Pre-commit what you'll do in a crash, before the crash decides for you.BEHAVIOR
Panic guardrail — what history says about drawdowns
The biggest retail loss is selling in a crash. Here is what every S&P 500 drawdown since 1927 actually did next.
Down −0.7% from the high. Read the table — this is normal, and selling locks the loss in.
All-time high 7,609.78 (2026-06-02)
After the S&P had fallen this far (1927–2026)
| Drawdown | Times | Median recovery | Avg next 12m | Positive next 12m |
|---|---|---|---|---|
| −10% | 17× | 23mo | +8.3% | 69% |
| −20% | 10× | 34mo | +8.0% | 67% |
| −30% | 6× | 73mo | +7.4% | 62% |
| −40% | 4× | 86mo | +7.4% | 59% |
Pre-commitment
I will not sell my NISA core in a drawdown. Even down −20/−30%, the next 12 months averaged about +8% and were positive ~2 of 3 times. Selling turns a paper loss into a permanent one and forfeits tax-free NISA quota I cannot rebuy.
Historical context, not a buy/sell signal and not advice. Price-return, dividends excluded; the future can differ.
FINDING08Where do those guardrails sit right now?Trigger lines plotted against the current price.CHART
After a crash, the next year was positive — every time
Average S&P 500 return over the 12 months AFTER the index had already fallen to each depth (1927–2026). Even down −30% or −40%, the next year averaged about +7–8% and was positive roughly 2 of 3 times. Panic-selling locks the loss in and forfeits the recovery. History, not a forecast. Not advice.
FINDING09How does writing decisions down make them better?The smallest decision journal that still kills hindsight bias.PLAYBOOK
Decision journal — log the urge before you act
Every impulse to sell, time, or chase goes here first. Logging adds a cooling-off step and a record you can hold yourself to.
Before you act — what the evidence says
Drawdowns recover: even down −20/−30%, the next 12 months averaged ~+8% and were positive ~2 of 3 times. Selling locks the loss in and forfeits NISA quota you can't rebuy.
No entries yet. Next time you feel the urge, log it here first.
Decision-support, not advice. Entries stay in your browser (localStorage) — nothing is sent anywhere.
FINDING10Does your strategy survive its own costs and taxes?The five-minute checklist to run before you believe any backtest.CHECKLIST
Cost & tax minimiser — the edge you actually control
You can't reliably beat the market, but you can stop bleeding to fees, tax and turnover. These are the levers that are 100% in your hands.
- ✓
Hold the cheapest vehicle.
VOO/VTI ~0.03%, オルカン ~0.06%. You can't beat that by hand — DIY replication costs more in spreads, FX and tax than the fee it saves.
- ✓
NISA before any taxable account.
Gains and domestic tax are zero inside NISA. Fill the ¥3.6M/yr quota first — it's the single biggest free edge you have.
- ✓
Don't churn.
Every trade pays spread; outside NISA every sale realises 20.315% tax. Turnover was the structural opponent that beat every active strategy we tested.
- ✓
One core, no satellite.
Value, quality, momentum, buyback and sector tilts all lost to the broad index after cost. A satellite only adds risk and fees.
- ✓
Don't hold cash waiting for a dip.
Front-load beat DCA 67% of years; buy-the-dip lost. Idle cash is a guaranteed drag against an uncertain dip.
Decision-support, not advice. General cost/tax structure for a JP retail NISA investor.