Swedroe: Time-Series Momentum Persists

November 06, 2017

As my co-author Andrew Berkin and I present in our book, “Your Complete Guide to Factor-Based Investing,” there is a large body of academic research demonstrating that, similar to some better-known factors like size and value, time-series momentum historically has recorded above-average excess returns.

Time-series momentum, also called “trend momentum” or “absolute momentum,” is measured by a portfolio long assets that has had recent positive returns and short assets that have had recent negative returns.

Compare this to the traditional (cross-sectional) momentum factor, which considers recent asset performance only relative to other assets. The academic evidence suggests that inclusion of a strategy targeting time-series momentum in a portfolio improves that portfolio’s risk-adjusted returns. Strategies that attempt to capture the premium offered by time-series momentum are often called “managed futures,” as they take long and short positions in assets via futures markets.

As we explain in our aforementioned book, to have confidence in the likelihood of a factor enduring into the future, there should be strong evidence of its persistence across long periods of time, its pervasiveness across asset classes as well as geography, its robustness to various definitions, its implementability (meaning it survives transaction costs), and that there are risk-based or behavioral explanations for its existence. In terms of both its persistence and pervasiveness, the time-series momentum factor has been documented across many stock markets and asset classes.

Momentum In Government Bonds

Adam Zaremba contributes to the literature on time-series momentum with his study “Performance Persistence of Government Bond Factor Premia,” which appears in the August 2017 issue of Finance Research Letters.

In it, he provides evidence of the pervasiveness of time-series momentum by examining its performance in international government bond markets. His investigations were based on 23 bond factor strategies related to volatility (adjusted duration, beta, idiosyncratic volatility, standard deviation, value at risk, downside deviation, and duration-times-yield), credit risk (an average of cross-sectional z-scores of budget deficit-to-GDP and net debt-to-GDP ratios, sovereign risk, credit rating, and total market value of debt), value (yield-to-maturity, term premium, term relative value, credit relative value, term and credit relative value, and 48-month yield change), and momentum (12-month price momentum, six-month yield momentum, 12-month yield momentum, six-month moving average, and 12-month moving average).

Zaremba tested these bond factor strategies in a sample of data from total return bond indexes from 25 countries over the period 1992 through 2016. The indexes are determined separately for five maturity buckets: 1-3 years, 3-5 years, 5-7 years, 7-10 years and more than 10 years, resulting in his investigation of a total 125 international government bond buckets.

Next, Zaremba ranked each of the bond buckets on returns and formed zero-investment, value-weighted portfolios that were long (short) in the quantile of buckets with the highest (lowest) variable. To assure the robustness of his results, he used three types of quantiles (tertiles, quartiles and quintiles), different weighting schemes (value weighting and equal weighting), and split the period into two subperiods (before and after the 2008 financial crisis). All the portfolios were reformed on a monthly basis.

Following is a summary of his findings:

  • The returns on factor premiums display a strong and positive relationship with their past performance; the best-performing (worst-performing) factors continue to overperform (underperform).
  • The results hold for the majority of return measurement periods (e.g., from one to 10 years) and are not dependent on the precise portfolio construction approach.
  • The effect is robust to many considerations, including alternative portfolio formation techniques, subsample and subsection analysis, and various formation periods.
  • The effect is particularly strong for long-term formation periods ranging from five to 10 years.
  • The factors with the highest past returns (top) outperform the factors with the lowest past return (bottom). Long/short portfolios (T-B) that go long (short) the top (bottom) factors display positive and significant means of monthly returns equaling 0.30% to 0.33%, depending on the portfolio specification.

Zaremba also performed tests to help determine the source of the momentum premium. He found that the momentum effect stems from cross-sectional variation in expected returns on particular factor premiums or styles due to slow investor learning and changes in liquidity (underreaction) rather than investor overreaction (performance-chasing).

Summary

As an investment style, time-series momentum has existed for a long time. Zaremba’s study provides strong out-of-sample evidence beyond the substantial evidence that already existed in the literature. As to reasons why we should expect trends to continue, the research suggests the most likely candidates include investors’ behavioral biases, market frictions, hedging demands, and market interventions by central banks and governments.

Such hedging programs and market interventions are still prevalent, and investors are likely to continue to suffer from the same behavioral biases that have influenced price behavior in the past.

Larry Swedroe is the director of research for The BAM Alliance, a community of more than 140 independent registered investment advisors throughout the country.

Find your next ETF

CLEAR FILTER