The strategy · step by step

How it works

The rule of the game: keep what is rising, take shelter when everything falls — and the figures that prove it.

Portfolio Tester backtests tactical asset-allocation strategies. The production strategy it runs is a 70/30 composite dual-momentum portfolio, rebalanced monthly.

Dual momentum combines two filters. Relative momentum ranks a small set of risk assets and holds the strongest. Absolute momentum keeps that position only while it outperforms a safe baseline; when momentum turns negative, the sleeve rotates to its defensive asset — plain cash, short-term T-bills (BIL) — rather than riding the asset down. The defensive asset used to be a gold-and-cash blend; it is now cash, because stepping out cleanly is what protects in the regimes that matter, while a gold parachute is a bet on a particular regime (see the long-term tests).

The composite splits capital 70% into a US equity-style sleeve (large-cap growth, value, blend) and 30% into a commodity-sectors sleeve. Each sleeve runs its own dual-momentum rotation; the two are recombined to fixed weights every month. Momentum is measured as a blend of 1-, 3- and 6-month lookbacks.

Where dual momentum comes from

Relative momentum is the older idea: among a set of assets, the recent winners tend to keep winning over the next few months — documented academically by Jegadeesh and Titman in 1993. On its own it always stays fully invested, dutifully holding the "least bad" loser through a bear market. Absolute momentum adds the missing gate: it compares an asset to its own trend and a safe baseline, and steps aside when that trend turns negative. Gary Antonacci combined the two in his 2014 book Dual Momentum Investing — pick the strongest asset, but hold it only if it also clears the absolute-momentum test. This site applies that rule monthly, per sleeve, against a market barometer (SPY for equities, DBC for commodities).

How the universes are built

Each universe is a small, complementary set spanning a single economic axis — US large-cap style (growth / value / blend), geography (US / developed / emerging), commodity sectors (precious, energy, agriculture, base metals), or bond credit quality (government / investment-grade / high-yield). The members are deliberately non-overlapping so that different ones lead in different regimes — which is exactly the rotation relative momentum is built to capture. The defensive (out-of-market) asset is always drawn from a different class than the sleeve it protects, so the fallback isn't correlated with what it's replacing.

The crypto universes are the exception to this shape. Bitcoin alone is not a set — it is a single, extremely volatile asset, so the rotation collapses to a pure in-or-out call: hold Bitcoin while its own momentum beats cash, step aside into cash otherwise. Bitcoin/Ethereum adds a second risk asset, turning that filter into a genuine two-way rotation. Both keep cash as the defensive and both are an exploratory, backtest-only testbed, never traded — the crypto section covers what they show and why their momentum window is shorter.

Finding the champions

Each asset class was tested by replaying the 2016–2026 history under different dual-momentum settings — a series of targeted, hypothesis-driven sweeps (not an exhaustive grid, and not a Monte-Carlo simulation) — ranked by return per unit of worst drawdown (CAGR ÷ maximum drawdown). The best configuration in each class is its "champion." The production portfolio combines the two champions favoured here — US equity-style and commodity-sectors, weighted 70/30 — while deliberately leaving bonds out, a macro-thesis choice rather than a performance one (see Theses). Combining the two classes did more than add return: it cut the worst drawdown materially versus either sleeve alone, which is the diversification the composite is built for. Over the 2016–2026 sweep window the deployed 70/30 — stepping out into cash — compounds at about 16% a year with a worst drawdown near −10% — roughly the S&P 500's return over the same window at half the drawdown.

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

These results come from one historical window. They show the strategy executed well in sample; they are not evidence that the edge survives across market regimes. A genuine out-of-sample or resampling test — including a true Monte-Carlo / bootstrap of returns — remains future work.

The Long-term page asks whether the logic survives across a century of regimes, on long proxy series. Here is the complementary check, on the actual US instruments over the window each one exists — the equity ETFs back to 2000 (opening at the dot-com peak, an unflattering start), the commodity sectors back to about 1992, and the composite to about 2001. The signal steps aside into cash, never into a commodity barometer, so the broad commodity index shown here is only a comparison line, not part of the rule. The two legs behave as the method predicts. This is a fidelity check, not a second claim of robustness. Each chart adds an underwater drawdown panel below the equity curve.

And the two legs combined — the deployed 70/30, rebalanced monthly between them, not a static 70/30 of the curves above:

The defensive asset — what the strategy holds when it steps out of the market — is the real diversification lever, more than the geography or style of the risk assets it rotates between. That lever is now plain cash: the edge in the regimes that matter comes from stepping out cleanly, not from a clever safe-haven. Get the exit right and the whole portfolio behaves better.

The investable universe

The live 70/30 strategy only ever holds seven of these (IWF/IWD/IWB and GLD/DBE/DBA/DBB), turning defensive into cash (BIL); SPY and DBC are read-only barometers. GDX is no longer held in production but stays selectable in the backtester. The rest make up the broader universe you can experiment with.

The monthly signal and the closing-auction window

The strategy decides on month-end closing prices — but that bar only exists at the close. Waiting for it to be confirmed would push any trade to the next morning's open, exposing the portfolio to overnight gaps and to a price the backtest never used. So a few minutes before the monthly close, the signal is computed on a provisional month-end bar built from the live intraday snapshot. That points to the assets the strategy will almost certainly want, and lets you submit Market-On-Close (MOC) orders that execute at the official closing price — the exact price the backtest assumes. Brokers and exchanges accept MOC orders up to 10 minutes before the close (3:50 PM ET), so the workflow fits inside that window.

The trade-off, stated plainly: the provisional signal can still flip in the final minutes. The strategy accepts this on purpose — trading at the close on a possibly-imperfect signal beats trading the next morning on a certain one, because over years, matching the backtest's execution price matters more than the rare last-minute flip.

How a European investor replicates all of this — UCITS equivalents, life-insurance mappings, access to US ETFs — lives in Investing from Europe. The exploratory Bitcoin/Ethereum testbed has its own page: Crypto.