Dual momentum: the rules, the evidence, and a backtest

Dual momentum is a monthly rule with two parts. Relative momentum picks the asset that has performed best over the past twelve months. Absolute momentum asks whether that asset has actually gone up over the same period, and moves to Treasury bills if it has not. Put together, the strategy holds the strongest trending equity market when trends are positive and steps aside when they are not. The rule fits in a sentence, and the sentence is the prompt at the bottom of this page.

What are the exact rules?

The version tested here uses three exchange-traded funds: SPY for US large-cap equities, EFA for developed international equities, and BIL for one-to-three-month Treasury bills. On the last trading day of each month:

1. Compute the 12-month total return of SPY and EFA.
2. Select whichever is higher. This is the relative momentum step.
3. If the selected fund's 12-month return is below zero, hold BIL instead. This is the absolute momentum step.
4. Hold that single position until the next month end.

That is the whole strategy. There is no volatility targeting, no partial weights, and no discretion. In any given month the portfolio is 100 percent in one of three funds.

Gary Antonacci introduced the approach in his 2012 paper Risk Premia Harvesting Through Dual Momentum and expanded it in the 2014 book Dual Momentum Investing. His flagship implementation, Global Equities Momentum (GEM), uses the same structure with an aggregate bond fund in place of T-bills for the defensive position. The T-bill version here is the more conservative reading of the rule and avoids the question of whether long bonds count as "safe."

Why might it work?

Each half of the rule rests on a separate body of evidence.

Relative momentum is one of the most replicated findings in empirical finance. Jegadeesh and Titman documented in 1993 that stocks with high returns over the past 3 to 12 months continued to outperform over the following months, and the effect has since been found across countries, asset classes, and time periods, including the 200-year sample in Geczy and Samonov's Two Centuries of Price Return Momentum. The usual explanations are behavioral: investors underreact to new information at first and then overreact, so trends persist longer than a random walk would predict.

Absolute momentum, also called time-series momentum, is the observation that an asset's own past return predicts its future return. Moskowitz, Ooi, and Pedersen's 2012 paper Time Series Momentum found this across 58 futures markets. The practical consequence is that when an equity index has fallen over the past year, its expected return over the next few months has historically been lower and its volatility higher. Stepping aside during those periods is what cuts drawdowns.

The reason to combine them is that each fixes a weakness of the other. Relative momentum alone will hold the "least bad" equity market in a global bear market, which is still a bear market. Absolute momentum alone, applied to a single index, misses the gains from rotating into whichever market is leading. Together, the strategy participates in whichever equity market is strongest and exits when neither is rising.

When does it fail?

Every trend-following rule has the same failure mode, and dual momentum is no exception: it needs trends to persist long enough to profit from them. Three environments have historically been painful.

Sharp V-shaped reversals. The rule takes up to a month to notice a decline and up to a month to notice the recovery. In a crash that reverses within a few weeks, the strategy can sell near the bottom and buy back higher. March 2020 is the canonical example; the 12-month signal for SPY turned negative at the end of March, after most of the decline, and turned positive again by the end of May, after a large part of the recovery.

Choppy, trendless markets. When the two equity funds trade places month after month and neither has a strong trend, the strategy pays transaction costs and taxes for switching while capturing little. Extended sideways periods, like 2015 to 2016, are the strategy's worst relative environment.

Long bull markets. Absolute momentum occasionally triggers a false exit during a correction inside a bull market, and the strategy then sits in T-bills while equities recover. Over a decade like the 2010s, a few of these can leave dual momentum well behind buy-and-hold US equities, even if the drawdown profile is better. An investor who benchmarks to the S&P 500 may find that gap harder to tolerate than the drawdowns it prevents.

The historical record on drawdown is the strongest argument for the strategy. The historical record on relative performance during bull markets is the strongest argument against it. Neither record is a guarantee of the next one.

What does a backtest show?

The card below is a live tear sheet of an ENSEMBLE model built from the prompt at the bottom of this page. The metrics are computed by the platform from daily closing prices, net of a simulated transaction cost on each trade, and refresh nightly.

A few things to look for on the full tear sheet:

• The drawdown chart against SPY. The strategy's deepest declines have historically been shallower than the equity market's, which is the point of the absolute-momentum filter. Look at 2008 and 2022 in particular.
• The holdings over time. You will see long stretches in a single fund and clusters of switches in choppy periods. The clusters are where the costs are.
• Rebalances per year. The model's run metadata records how often the position actually changed. A rule evaluated monthly does not trade monthly; it trades when the signal changes, which has historically been a few times a year.

For definitions of each figure, see backtest metrics explained. For the ways a backtest like this one can mislead, see common backtesting mistakes, especially the sections on short samples and on the cost of whipsaws.

How to run it and change it

The prompt below builds this exact model. Run it as written, or change one thing and run it again.

Variations worth testing, each a small edit to the sentence:

• Lookback. Replace "12-month" with "6-month" or "10-month". Shorter lookbacks react faster and trade more.
• Defensive asset. Replace BIL with AGG or IEF to test the GEM-style bond version. Note that this adds duration risk to the defensive position, which mattered in 2022.
• Universe. Add EEM for emerging markets, or replace EFA with VEA. More candidates raise the chance of catching the leading market and the chance of overfitting.
• Rebalance timing. Some implementations evaluate the signal weekly and act only on month end; others act whenever the signal changes. Describe the timing you want and compare.

Each variation is a new public model with its own tear sheet, so you can compare them side by side rather than reasoning about the difference.

Practical notes

Dual momentum is a taxable-account-unfriendly strategy in most years, because positions are often held less than twelve months. Investors who run it typically do so in tax-advantaged accounts.

It is also sensitive to the signal date. Two investors running the same rule on the last and first trading days of the month can hold different funds for weeks at a time. The published research uses month-end; so does this model.

Finally, the strategy holds one fund at a time. That concentration is intentional, since diversification within the position would dilute the momentum signal, but it means the portfolio inherits the full volatility of whichever fund it holds. Investors who want the drawdown protection with less concentration often pair dual momentum with a fixed core allocation, which is exactly what an ENSEMBLE portfolio is for: hold 70 percent in a 60/40 core and 30 percent in this strategy, rebalanced quarterly.

Backtests are hypothetical, past performance does not guarantee future results, and ENSEMBLE is research software rather than an investment adviser.