Momentum & factors

Momentum on dual momentum portfolios: rotating between whole strategies

Momentum & factors · Published 2026-10-22

Dual momentum is a simple rule-based approach to asset allocation: own what has been rising relative to the alternatives, but only if it has also been rising in absolute terms. A natural extension is to apply the same logic one level higher, treating whole dual momentum portfolios as the assets and rotating between them. This article explains both ideas, and why the second one, while appealing, needs especially careful testing.

Dual momentum in brief

The term was popularised by Gary Antonacci, first in a 2012 working paper on risk premia harvesting and then in his 2014 book Dual Momentum Investing. The idea combines the two kinds of momentum described in our momentum overview:

• Relative momentum chooses among candidate assets, for example U.S. stocks versus international stocks, by picking the one with the higher return over the look-back period, usually twelve months.
• Absolute momentum then checks whether the chosen asset's return over that period beat a risk-free rate such as Treasury bills. If it did, the strategy holds the asset; if not, it moves to a defensive asset such as short or intermediate-term bonds.

The well-known "global equities momentum" version rotates monthly between a U.S. stock index, an international stock index and bonds. The absolute filter is what does most of the work in bear markets: by stepping aside when equities have had a negative year, the rule historically avoided much of the worst of long declines, at the price of lagging during sharp rebounds and paying for occasional false signals.

Applying momentum to a country basket

The same machinery can be applied to a broader set of assets. Country exchange-traded funds are a natural candidate: providers such as iShares list single-country equity funds covering dozens of markets, which lets a researcher test rotation across countries using liquid instruments. A typical country rotation ranks the funds by twelve-month return, holds the top few, and applies an absolute momentum filter to move to bonds or cash when even the leaders are falling.

Two features make country baskets interesting. First, national markets still have distinct sector mixes, currencies and economic cycles, so returns across countries diverge enough for relative momentum to find leaders. Correlations between developed markets have risen since the 1990s, but they remain well below one, especially between developed and emerging markets. Second, cross-country momentum is supported by the academic evidence: Asness, Moskowitz and Pedersen (2013) found momentum in equity country indices as well as in individual stocks.

The practical limits are also clear. Single-country funds have higher fees and wider spreads than broad index funds, their histories are often short, and some markets have small, concentrated indices dominated by a few companies.

One level up: momentum on portfolios

Suppose a researcher has two different dual momentum portfolios, for example a country rotation and a multi-asset rotation across stocks, bonds, real estate and gold. Each produces a monthly return series. The extension is to treat these two series as assets and apply momentum to them: each month, hold whichever portfolio has the better recent performance, perhaps with its own absolute filter.

The intuition is that strategies, like assets, go through periods when their logic suits the market and periods when it does not. If those periods persist, a rule that follows the recently stronger strategy should capture some of that persistence. There is some support for the idea in research on factor momentum: Ehsani and Linnainmaa (2022) found that factor returns are positively autocorrelated, and that a strategy timing factors by their own past returns captured much of the profitability of individual stock momentum.

Why strategy rotation is easy to overfit

The extension multiplies the number of choices a researcher makes: the assets in each underlying portfolio, the look-back periods at both levels, the number of holdings, the defensive asset, and the rebalancing date. Each choice that was made after looking at the data reduces the evidence value of the final backtest. Several points deserve attention.

• Short histories. Many ETFs began trading in the late 1990s or 2000s. A rotation tested over fifteen or twenty years may include only one or two full market cycles, and its result may depend on a handful of switches.
• Selection of the components. If the two underlying portfolios were themselves chosen because they backtested well, the top-level result inherits that selection bias. Bailey and López de Prado's "deflated Sharpe ratio" was designed to adjust for this kind of multiple testing.
• Turnover and taxes. Switching whole portfolios means selling everything at once. Costs, slippage and, for taxable investors, realised gains can be material.
• Look-ahead in timing. A signal computed on the last trading day of a month cannot be traded at that same close without bias. Using the next day's prices for execution is a simple safeguard.

A disciplined way to test the idea

1. Fix the rules of the underlying portfolios in advance, ideally from published sources, rather than tuning them.
2. Use the longest data available, including index data before ETFs existed, while noting that index returns ignore fees.
3. Hold out a final period that is not looked at until the rules are frozen.
4. Report the number of variations tried and the spread of their results, not only the best one.
5. Check whether the rotation adds anything beyond a simple equal-weight mix of the two portfolios rebalanced monthly. Diversifying between strategies is often a tougher benchmark than it seems, as our article on combining diversified alpha explains.

Takeaways

Dual momentum is a transparent rule with a long history of study, and its absolute filter is a clear way to reduce exposure during prolonged declines. Applying momentum to whole portfolios is a reasonable hypothesis with some theoretical backing from factor momentum research. But the extension sits on top of many design choices and short data, so the burden of proof is high. Readers interested in why momentum strategies fail when they do should continue with momentum crashes.