Finding a Better Safe Haven via U.S. Treasuries Dual Momentum

Does a dual momentum selection/weighting approach applied to the U.S. Treasuries term structure identify a safe haven superior to any one duration? In his February 2015 paper entitled “The Search for Crisis Alpha: Weathering the Storm Using Relative Momentum”, Nathan Faber tests a dual momentum safe haven based on U.S. Treasuries of different durations as proxied by either constant maturity indexes or exchange-traded funds (ETFs). He constructs constant maturity indexes from 1-year, 3-year, 5-year, 7-year, 10-year and 20-year constant maturity U.S. Treasuries yields by each month accruing a coupon and repricing at the new yield. For ETFs, he uses total returns for five iShares U.S. Treasuries ETFs: SHY (1-3 years), IEI (3-5 years), IEF (7-10 years), TLH (10-20 years) and TLT (20+ years). The dual momentum approach consists of the following steps:

1. Calculate the return from 10 months ago to one month ago for each duration.
2. Subtract from the return of each duration that of 1-year U.S. Treasuries (SHY) if using constant maturity indexes (ETFs) to calculate an excess return as a measure of intrinsic (absolute or time series) momentum.
3. Discard any durations with negative excess returns.
4. Rank remaining durations based on risk-adjusted excess returns, with variances used to indicate risk, as a measure of relative momentum and assign weights to these durations based on their ranks. If no durations have positive excess returns, assign 100% weight to 1-year U.S. Treasuries (or SHY if using ETFs).

He then investigates the performance of this dual momentum strategy as a safe haven during S&P 500 crises defined in two ways: (1) drawdowns of at least 20% peak to trough; or, (2) monthly declines of at least 5%. He further tests a specific strategy that is long the S&P 500 Index (or SPY if using ETFs) when above its 10-month SMA (SMA10) and in either the dual momentum safe haven portfolio or in a fixed duration (1-year or 20+ years) when below its SMA10. Using data for the yields/indexes/funds specified above since 1962 for constant maturity index tests and since 2003 for ETF tests, all through 2014, he finds that:

• Compared to equal weight (EW) across durations and intrinsic momentum based on a 9-month lookback interval alone, respectively, the dual momentum safe haven portfolio based on constant maturity indexes during 1962-2014 has:
  • Higher gross annualized return (7.29% versus 6.60% and 7.26%).
  • Lower annualized volatility (5.04% versus 5.76% and 5.42%).
  • Higher gross Sharpe ratio (0.32 versus 0.16 and 0.29).
  • Shallower maximum drawdown (5.90% versus 10.68% and 7.19%).
• Sensitivity tests for the constant maturity indexes during 1962-2014 indicate that:
  • A 9-month lookback interval (with a skip-month) is near optimal in the range six to 12 months, but three of these seven lookback intervals do not perform well for Crisis Measure 1.
  • Including the skip-month in momentum calculations generally improves results. 
• Compared to 1-year Treasuries and 20-year Treasuries, respectively, using the dual momentum safe haven portfolio based on constant maturity indexes within the S&P 500 Index SMA10 timing strategy during 1962-2014 results in:
  • Higher or comparable gross annualized return (9.7% versus 9.0% and 9.7%).
  • Lower or comparable annualized volatility (10.9% versus 10.6% and 12.2%).
  • Higher Sharpe ratio (0.21 versus 0.15 and 0.19).
  • Higher maximum drawdown (25.3% versus 24.0% and 23.6%).
• Compared to SHY and TLT, respectively, the dual momentum safe haven portfolio based on U.S. Treasuries ETFs for months when the SPY SMA10 timing strategy is out of SPY during 2003-2014 has:
  • Middling gross annualized return (6.34% versus 3.02% and 8.86%).
  • Moderate annualized volatility (9.84% versus 1.89% and 20.16%).
  • Sharpe ratio higher than that for TLT (0.34 versus 0.29).

In summary, evidence suggests that constructing a safe haven portfolio by applying dual momentum across the U.S. Treasuries term structure may have some advantages over picking a single duration as safe haven from U.S. equity market crises.

Cautions regarding findings include:

• Overall, improvements offered by the dual momentum safe haven portfolio appear modest relative to potential trading frictions and snooping bias. Specifically:
  • Constant maturity indexes are not investable. Incorporating costs of maintaining liquid tracking funds for these indexes would reduce their returns.
  • Portfolio return calculations are gross, not net. Costs of monthly rebalancing of duration allocations would reduce returns of the dual momentum safe haven portfolio.
  • Choices of lookback interval length, excess return threshold, risk-adjustment metric, crisis specification and equity market timing metric may incorporate direct or borrowed data snooping bias, thereby overstating expectations.
  • Lookback interval sensitivity testing considers only intervals of six to 12 months. Findings in “Simple Term Structure ETF/Mutual Fund Momentum Strategy” suggests shorter intervals may work better.
• S&P 500 Index analyses ignore dividends, thereby understating performance of associated strategies.
• The dual momentum safe haven portfolio based on U.S. Treasuries may not work for portfolios with material positions in assets other than U.S. equities (which may have different crises).
• The 2008 financial crisis dominates the relatively short tests based on ETFs and may not be representative of future conditions.

See also Simple Debt Class Mutual Fund Momentum Strategy?.

Frequently Asked Questions

What is a dual momentum safe haven portfolio built from U.S. Treasuries?

A dual momentum safe haven portfolio is a Treasury allocation that combines absolute (intrinsic) momentum and relative momentum across bond durations, such as 1-year to 20+ year maturities. It discards durations whose recent excess return over 1-year Treasuries is negative, then weights the rest by rank of risk-adjusted excess return. If no duration has a positive excess return, it holds 100% in 1-year Treasuries (or SHY for ETFs).

Does dual momentum across the Treasury term structure beat picking a single bond duration as a safe haven?

Dual momentum across the Treasury term structure modestly outperformed fixed-duration and equal-weight alternatives in 1962-2014 tests using constant maturity indexes. It delivered a 7.29% gross annualized return versus 6.60% for equal weight, and a Sharpe ratio of 0.32 versus 0.16, with a shallower maximum drawdown (5.90% versus 10.68%). The improvements appear modest relative to trading frictions and data snooping bias.

How is a dual momentum Treasury safe haven calculated and tested against stock market crises?

The dual momentum Treasury safe haven is calculated monthly using each duration's return from 10 months ago to one month ago, minus the return of 1-year Treasuries (SHY for ETFs). Durations with negative excess returns are dropped and the rest are ranked by risk-adjusted excess return, using variance as risk, to set weights. It was tested during S&P 500 crises defined as drawdowns of at least 20% or monthly declines of at least 5%, and within a strategy that holds the S&P 500 when above its 10-month SMA and the safe haven when below, using 1962-2014 index data and 2003-2014 ETF data.

How does a dual momentum Treasury safe haven compare with holding 1-year or 20-year Treasuries when stocks are below their 10-month moving average?

In the S&P 500 10-month SMA timing strategy during 1962-2014, the dual momentum safe haven had a higher or comparable gross annualized return (9.7% versus 9.0% for 1-year and 9.7% for 20-year Treasuries). Its Sharpe ratio was higher (0.21 versus 0.15 and 0.19), but its maximum drawdown was also higher (25.3% versus 24.0% and 23.6%). In the 2003-2014 ETF test, its Sharpe ratio exceeded that of TLT (0.34 versus 0.29), with middling return and volatility between SHY and TLT.

What are the limitations and risks of using a dual momentum Treasury safe haven strategy?

The dual momentum Treasury safe haven's improvements are modest relative to trading frictions and data snooping bias, since returns are gross and monthly rebalancing costs would reduce them. Constant maturity indexes are not investable, and choices of lookback interval, threshold, and risk metric may overstate expectations. The results ignore S&P 500 dividends, may not hold for portfolios with significant non-U.S.-equity assets, and the ETF tests (2003-2014) are dominated by the 2008 financial crisis, which may not be representative of future conditions.