Out-of-sample · 1926–2025

100-year proof

The strategy replayed over a century of data: what holds, and what depends on the era.

The champion sweeps were run on a single window, 2016–2026 — a calm, gold-friendly decade. This page does the opposite: it replays the strategy across a full century, 1926–2025, through the Depression, the 1940s, the 1970s and every modern crisis, to ask whether the edge is structural or an artefact of the sample.

The long series are reconstructions, run through the same validated engine that powers the live backtest — never recomputed by a separate chart script. Equities use Fama-French growth/value (book-to-market); commodities use producer-price indices. These proxies are faithful to the logic, not to the exact ETFs the portfolio trades — so what follows validates the theory the strategy rests on, not its precise returns. The deployed instruments, over the recent window, live in the How it works section.

The equity engine: the filter, then the rotation

The equity engine has two moving parts, and this first figure takes them apart over the full 1926–2026 Fama-French history, with cash as the only parachute so gold plays no role. Three lines: the passive US index (the total Fama-French market); that same index run through the absolute-momentum filter, stepping into cash on a sell signal; and the growth/value rotation that holds the strongest of growth, value and the market while invested, through the identical filter. Because the filter is the same, lines (2) and (3) step into cash on the same 384 months — verified identical — so the gap (3) − (2) is the rotation and nothing else, while (2) − (1) is the filter.

The split is revealing. The filter (2 vs 1) is pure risk control: it cuts the worst drawdown from −84% to −46% but costs about 0.8 points of annual return — stepping aside is not free. The rotation (3 vs 2) is the opposite trade: it adds about 1.7 points of CAGR, compounding to roughly 4.8× the terminal wealth of the plain index-DM over the century — but it deepens the drawdown, from −46% to −59%. Over-weighting whichever style is strongest is exactly what loads up on the style about to crack hardest, and in the 1929–1935 collapse that cost it: the lower panel — the cumulative (3) ÷ (2) — shows the rotation contribution distributed across almost every decade (the 1940s through the 2020s are all positive), with one glaring exception, the 1930s, where it gave back about 30%. That single episode is why, on the full window, the rotation's return-per-unit-of-drawdown ratio (0.19) sits slightly below the plain filter's (0.21) even though it makes far more money. The rotation earns its keep on return, paid for in tail risk.

All figures are month-end closes; daily-marked drawdowns run deeper — see the note on granularity in the Guide.

The parachute: cash or gold?

The rotation above stepped aside into cash. Should the parachute hold gold as well — or is plain cash enough? To answer it on the deepest possible history we again use the Fama-French research data back to 1926, running the same growth/value rotation — hold the strongest of growth, value and the broad market by momentum, step aside when absolute momentum turns negative — and compare two parachutes for that step-aside: cash, and a 50/50 gold-and-cash blend. The passive market index is the reference. Gold was a fixed, illegal-to-hold price until 1971, so the 50/50 parachute holds only cash before then — which is why the two strategy lines are identical until 1971 and split only afterwards.

The verdict is the one cash keeps earning across this page. The worst drawdown of the entire century is identical for both parachutes — −59%, in the Great Depression aftermath — because the worst tail landed when gold could not be held at all, and rotation alone (not the parachute) is what cushions it to −59% against the market's −84%. After 1971, when gold is finally in the blend, it does add return: the 50/50 leg compounds at 12.4% against cash's 11.6%. But it pays for that with a deeper worst drawdown — −26.8% versus cash's −24.7% — and a slightly worse return-per-unit-of-drawdown ratio. Gold flatters the upside and creeps the downside; it does not buy tail protection. The lower panel makes the point directly: through the worst stretches the two drawdown lines sit on top of each other or the gold line runs deeper, never shallower. Cash is the robust floor — so the deployed parachute is plain cash; gold's place, if any, is static self-custody held outside the engine, not the defensive sleeve the strategy rotates into.

The commodity sleeve

The third sleeve is commodities, and it too can be pushed back to 1926 — though only on a coarser input: the producer price indices (PPI) for energy, agriculture, base metals and the broad all-commodities aggregate, plus the same century gold series the parachute chart uses. This runs the exact deployed commodity-sectors leg — hold the strongest of gold, energy, agriculture and base metals by 1-3-6 momentum (top-1, Antonacci-style absolute filter), step into cash on a sell signal — over nearly a full century, against buy-and-hold of the broad commodity index.

Over 1926–2025 the dual-momentum leg compounds at 7.6% with a −31% worst drawdown, against buy-and-hold's 2.8% at −40%, and it sits in cash about a fifth of the time. The compound-return gap is flattered and should not be over-read: the broad PPI barometer is a smoothed producer-price series that barely moves (≈2.8% a year, essentially commodity inflation), and PPI levels are far smoother than tradable futures, so almost any trend-follower towers over them. The meaningful read is the lower panel: by stepping into cash through the worst commodity busts (shaded), the sleeve keeps its worst drawdown (−31%) shallower than buy-and-hold's (−40%) — the same step-aside behaviour the other sleeves show. Before 1971 — when gold was a fixed, un-investable price — the rotation is effectively energy/agriculture/metals, with gold selected only briefly (around the early-1930s commodity collapse, the 1933/1934 dollar revaluations, and the late-1960s as gold began to float).

The 70/30 composite — diversification across the century

This is the payoff. Combine the two century legs — the growth/value equity leg and the commodity-sectors leg, both with a plain cash parachute — into a single 70/30, rebalanced monthly by the validated engine (not a blend of finished curves), over 1926–2025. The diversification holds, and bluntly: the two legs are almost uncorrelated (monthly correlation ≈ 0.01), so the composite's worst drawdown (≈ −46%) lands shallower than its riskiest leg — the equity leg at ≈ −59% — and far shallower than the −84% of plain market buy-and-hold. The composite compounds at about 10.5% a year against the market's 10.3%, for roughly half the worst drawdown.

The honest caveat is the 1929–33 Depression, the focus of the lower panel. There the composite still falls about −44% — shallower than the equity leg's −58% and the market's −84%, but far deeper than the commodity leg's −14%, because 70% of the book is in equities and the cash filter does not catch the first leg down. Diversification softens the worst tail; it does not abolish it. And the levels here are not to be trusted — the commodity leg runs on smoothed producer-price data, so read the shape of the drawdowns and the gap between the legs, not the compound numbers.