Diversified Dual Momentum (Newfound Research)

Diversified Dual Momentum comes from Corey Hoffstein of Newfound Research, who in January 2019 showed how much the results of Gary Antonacci's Dual Momentum (GEM) depend on its lookback. Instead of a single model on 12-month returns, it runs seven copies of the GEM rules with lookbacks of six to twelve months and holds each in a seventh of the portfolio.

The idea#

Hoffstein called the problem specification risk. Momentum and trend have long records across many lookbacks, and there is no good reason to expect a 9-month model to do better or worse than a 10-month one. Yet over the same decade, GEM variants that differed only in lookback ended with very different results, and one could lose money in a year when its neighbor gained. Those gaps are luck, and luck does not even out later.

His answer was to stop betting on one lookback. Holding all seven versions at once keeps the style, dual momentum on US and non-US stocks with bonds as the fallback, while averaging away most of the luck of any single choice. The result should sit near the middle of the seven, with fewer sudden all-or-nothing switches.

How it works#

At the close of the last trading day of each month, for each lookback of 6, 7, 8, 9, 10, 11, and 12 months:

1. Compare the S&P 500's (SPY) return over the lookback with that of Treasury bills (BIL).
2. If US stocks beat bills, that seventh holds whichever of US stocks (SPY) and non-US stocks (VEU) had the higher return over the lookback.
3. Otherwise, that seventh holds US aggregate bonds (AGG).

Because each seventh decides on its own, the portfolio can hold one, two, or all three funds at once. Months are 21 trading sessions here, so the lookbacks run from 126 to 252 sessions. Newfound rebalanced the seven copies back to equal weight once a year; this version resets them every month.

What the backtest shows#

The ETF test runs from mid-2008. Diversified Dual Momentum compounded about 8% a year, less than the S&P 500 and roughly level with a 60/40 portfolio. Its worst drawdown, about 31%, was well short of the S&P 500's but no better than the 60/40's, and it trailed both on risk-adjusted return.

It protected well in 2008, gaining 6% over the rest of that year while the S&P 500 fell by a third. Choppy markets were harder: it made nothing in 2010, lost money in 2011, 2015, and 2018, and gained only 4% in 2020. Its worst drawdown came in the crash of February and March 2020, when all seven copies were in stocks; it recovered by January 2021. In 2022 it lost 18%, the same as the S&P 500, as both stocks and the bond fallback fell.

Since Newfound's post in early 2019 it has compounded about 10% a year. With simulated history the test starts in 1991 and looks stronger, at over 11% a year with a better risk-adjusted return than the S&P 500. It lost less than the index in 2000 and gained in 2001 and 2002 while stocks kept falling.

When it struggles#

• Sudden crashes. Monthly signals on six- to twelve-month returns cannot react to a decline that happens within weeks.
• Sideways markets. When trends keep reversing, several of the seven copies can switch at the wrong time together.
• Stocks and bonds falling together. Aggregate bonds offered no shelter in 2022.

Using it on Tactfolio#

The live strategy above runs the seven GEM copies side by side, rebalanced monthly. Copy it to change the range of lookbacks, use Treasury bills as the fallback, or compare it with the single-lookback Dual Momentum and the multi-asset Composite Dual Momentum.

Year by year

* Partial year.

The rules as implemented

This is the exact tree Tactfolio runs, rebalanced monthly with signals and trades at the close. Open it to inspect or copy it.

Sources and caveats

• Corey Hoffstein, Fragility Case Study: Dual Momentum GEM (Newfound Research, 2019)
• Gary Antonacci, Dual Momentum Investing (2014) and FAQ

• Each sleeve follows the GEM decision tree in the source: US stocks are first tested against Treasury bills, then compared with non-US stocks. Months are 21 sessions, so the lookbacks are 126 to 252 sessions.
• SPY, VEU, AGG, and BIL stand in for the S&P 500, world ex-US stocks, the Barclays US Aggregate, and Treasury bills, as in Tactfolio's single-lookback Dual Momentum.
• The source rebalanced the seven sleeves back to equal weight once a year; here they are reset to a seventh each at every monthly rebalance.
• Signals and trades use the close of the last trading day of each month, as in the source.

Common questions#

What is Diversified Dual Momentum?#

It is Newfound Research's version of Gary Antonacci's Global Equities Momentum that holds seven copies of the rules, each with a different lookback from six to twelve months, a seventh of the portfolio each.

How is Diversified Dual Momentum different from GEM?#

GEM makes one decision on 12-month returns and holds a single fund. The diversified version makes seven decisions on seven lookbacks, so it often holds a mix of US stocks, non-US stocks, and bonds, and it changes gradually rather than all at once.

What ETFs does Diversified Dual Momentum use?#

SPY for US stocks, VEU for non-US stocks, AGG for bonds, and BIL as the Treasury bill benchmark.

Why use several lookbacks instead of the best one?#

The best lookback in a backtest is largely luck and rarely stays best. Holding several spreads that luck, which Newfound found lowered the worst drawdown and kept returns near the middle of the range.

Related strategies

• Adaptive Asset Allocation (AAA)
• Dual Momentum: Antonacci’s Global Equities Momentum (GEM)
• Composite Dual Momentum
• GTAA Aggressive 3 (Faber GTAA AGG3)