Dual momentum (GEM): switching between stocks and bonds on 12-month returns

Dual momentum is a rule Gary Antonacci set out in a 2012 paper and his 2014 book Dual Momentum Investing. It uses two kinds of momentum together: absolute momentum, which checks whether an asset's own past return beat short-term Treasury bills, and relative momentum, which checks which of several assets rose more. The simplest form in the book is Global Equities Momentum (GEM), and in Korea it is presented as a leading example of dynamic asset allocation. This article runs GEM with ETFs and compares it with fixed-weight allocations. The figures are as of September 30, 2026 and were calculated by the operator.

Rules

• Signal: at the close on the last trading day of each month, compare SPY's (US stocks) return over the past 12 months with the return of 3-month Treasury bills over the same 12 months. The bill return compounds the 3-month bill rate (^IRX) day by day. The bill ETF BIL listed in 2007, after the start date, so it was not used.
• When in stocks: if SPY beat bills, hold whichever of SPY and EFA (developed markets outside the US) had the higher 12-month return.
• When in bonds: if SPY did not beat bills, hold AGG (the US aggregate bond market). The bonds' own momentum is not checked.
• Trading: when the signal changes, sell the whole holding and buy the new asset at the open of the next month's first trading day. A 0.15% one-way cost was charged on every traded amount and dividends were reinvested. Everything was sold at the close on September 30, 2026.
• Stocks outside the US: Antonacci's rule uses the MSCI ACWI ex US index, which includes emerging markets. ACWX, which tracks it, listed in March 2008 and the similar VEU in March 2007, so EFA (developed markets only), with data from 2001, was used instead. Results with ACWX are given separately below.
• Start: AGG listed in September 2003, so the first signal was taken at the end of September 2004, when a 12-month return could first be calculated, and $10,000 went in at the open on October 1, 2004.

Results (as of September 2026)

The comparisons are fixed allocations started on the same day and reset at the open of the first trading day of each January. 60/40 is SPY 60% and AGG 40%, and equal thirds simply holds the three assets GEM moves between at the same weight.

Over 264 months GEM held SPY for 142 months (54%), EFA for 79 (30%) and AGG for 43 (16%), and changed its holding 34 times. Each change meant two trades, a sale and a purchase, and costs added up to $3,326. With no costs the annual return was 9.1%, so costs took about 0.5 points a year. Of the 21 years from 2005 to 2025, GEM lost money in six and SPY in three. GEM had more losing years, but its largest yearly loss, −16.4% in 2022, was smaller than SPY's 2008 (−36.8%).

The max drawdown of −33.7% is the decline from February 19 to March 23, 2020, taken while holding SPY, and it equals SPY's drop over the same stretch. Looking only at the years from 2015, GEM and SPY had the same max drawdown, and GEM's annual return was 6.7 points lower.

Before and after the book

The book came out in November 2014. From October 2004 to December 2014, GEM returned 10.5% a year with a max drawdown of −23.5%, a higher return and a smaller drawdown than SPY (8.3% a year, −55.2%). From January 2015 to September 2026 the order flipped, with GEM at 7.0% a year and SPY at 13.7%, and GEM also trailed 60/40 (8.9%) and equal thirds (7.8%). Most of the earlier lead came from the single year 2008. The later period had no long, drawn-out decline like 2008; instead, short drops followed by quick recoveries came again and again, in 2015 to 2016, late 2018, 2020 and 2022.

2008, 2020 and 2022

• 2008: at the end of January 2008, SPY's 12-month return was −2.7%, below bills (4.15%), so GEM moved to AGG on February 1. It held AGG until the end of October 2009, a stretch in which SPY returned −21.8% and AGG +9.1%. GEM's 2008 was −2.4%. Still, it lost 13.6% holding EFA from its October 31, 2007 high to the end of January 2008, and the AGG it moved to was also 11.5% below its February 1 level on October 10, 2008, a day its market price fell 6.8%. From the high to October 10, 2008 the decline was −23.5%.
• 2020: the 12-month return through the end of February was positive, so GEM took the whole February 19 to March 23 drop in SPY. When SPY's 12-month return reached −6.9% at the end of March, it moved to AGG on April 1 and returned to SPY on June 1. In those two months SPY rose 22.4% and AGG 2.1%. It took the full decline and missed the rebound: GEM −1.9% for the year, SPY +18.3%. A 12-month rule checked once at month end cannot react to a crash that is over within a month.
• 2022: holding SPY from January to April, it fell 13.0%. At the end of April SPY's 12-month return was 0.0%, below bills (0.15%), so it moved to AGG on May 2, when AGG's own 12-month return was already −8.6%. GEM returned −3.9% from May to December and −16.4% for the year, not far from SPY (−18.2%). While it held AGG through the end of June 2023, SPY rose 9.8% and AGG fell 1.4%. In a decline driven by rising rates, bonds were no refuge.

Antonacci's figures and what research shows

• The 2012 paper: "Risk Premia Harvesting Through Dual Momentum" was posted on SSRN in April 2012 and won first place in that year's paper competition of the National Association of Active Investment Managers (NAAIM). It applied the same method to four pairs: US and foreign stocks, high-yield and credit bonds, equity and mortgage REITs, and gold and Treasuries. For 1974 to 2011 the equity pair returned 15.79% a year, above the 11.68% average of the two stock indexes (SSRN).
• GEM in the 2014 book: the book reports that GEM returned 17.43% a year from 1974 to 2013, with a standard deviation of 12.64% and a max drawdown of −22.7% (as summarized by CXO Advisory). These are index results that do not deal with trading costs or taxes. In this article's ETF calculation, GEM returned 8.6% a year from October 2004 with a max drawdown of −33.7%.
• Time-series momentum research: in a 2012 Journal of Financial Economics paper, Moskowitz, Ooi and Pedersen reported that across 58 futures markets, the sign of the past 12-month return tended to match the direction of the next month's return. Absolute momentum applies this property to a single stock market.
• The 12-month window: changing only the lookback in the same rule gave 6.0% a year for 6 months (68 changes), 6.1% for 9 months (51) and 8.6% for 12 months (34). The result leans heavily on the choice of window, so the 12-month result is hard to read as a property of the rule as a whole.
• The two kinds of momentum apart: using absolute momentum alone and moving only between SPY and AGG gave 9.5% a year with a max drawdown of −33.7% (14 changes). Using relative momentum alone and always holding the better of SPY and EFA gave 10.1% a year with −56.5%. Absolute momentum is what cut the drawdown, and over this period the relative momentum that picked EFA added no return.
• With ACWX: running the same rule from May 2009, once ACWX had 12 months of data, gave 8.8% a year for the ACWX version (27 changes) and 7.6% for the EFA version (33 changes). The results from 2004 are the EFA version.

Related names used in Korea

In Korea, Kang Hwan-guk's 2021 book The Giants' Portfolio presents dual momentum and Wouter Keller's rules as dynamic asset allocation. This article did not calculate the rules below and lists only their names and original sources.

• VAA (Vigilant Asset Allocation): a rule from the 2017 SSRN paper "Breadth Momentum and Vigilant Asset Allocation (VAA)" by Keller and Jan Willem Keuning. Instead of each asset's own trend, it moves to defensive assets based on how many of the offensive assets have negative momentum (breadth).
• LAA (Lethargic Asset Allocation): a rule from Keller's 2019 SSRN paper "Growth-Trend Timing and 60-40 Variations". It switches part of the portfolio only when both the US unemployment trend and the S&P 500 trend are bad, and keeps the rest at fixed weights.
• Kang-style combinations: the same book presents holding VAA, LAA and dual momentum in thirds. Details can differ between editions, so the original should be checked.

When the rule worked and when it did not

• It worked in long declines: in a decline that ran over several months, as in 2008, the 12-month return turned negative early and most of the fall was avoided. Most of the lead from October 2004 to December 2014 came from this.
• It was late for crashes that end within a month: in 2020 it moved to bonds after taking the whole decline and missed the rebound that followed right away.
• It had nowhere to hide when bonds fell too: the AGG it moved to in May 2022 kept falling. The rule does not check the bonds' own momentum.
• It earned less in long stretches led by US stocks: from 2015, while SPY rose 13.7% a year, GEM kept moving to EFA and back and returned only 7.0% a year.

Related concepts and search terms

• Absolute (time-series) momentum: deciding whether to hold an asset by its own past return. It belongs to the same family as rules that check whether the price is above a moving average; results for the 200-day and 10-month lines are in Moving-average trading.
• Relative momentum: picking whichever of several assets rose more. It is the same idea as the momentum factor used in stock selection.
• Dynamic or tactical asset allocation: changing weights by a preset rule. Fixed weights are covered in Asset allocation.
• Trend filters with monthly investing: results for selling below the 200-day line while investing every month are in the 200-day rule backtest.

Search terms: dual momentum, global equities momentum, GEM strategy, Antonacci, absolute momentum, relative momentum, tactical asset allocation, VAA, LAA, and the Korean terms for dual momentum and dynamic asset allocation.

Limits of these numbers

• The calculation covers one period, October 2004 to September 2026, and it contains only one long decline, 2008. A large part of the result rests on that one event.
• The signal is taken on a single day at month end. The same rule gave very different results when the lookback changed (6 months 6.0% a year, 12 months 8.6%).
• EFA stands in for the original rule's ACWI ex US. It leaves out emerging markets, so relative momentum may behave differently from the original rule.
• Every change trades the whole account. For Korean residents, gains on foreign ETF sales above ₩2.5 million a year are taxed at 22% (including local income tax), so each change can create tax, which this calculation leaves out.