Can Long-Only Investors Use Momentum to Beat the US Treasury Market?

The literature on momentum is vast. No previous study, however, has examined trading rules along government bond term structures. Under index-duration-neutral and long-only constraints and with low trading costs, an equally weighted average strategy across 20 look-back windows produces an information ratio of 0.46, given US Treasury total return data from March 1985 through December 2013. Unlike momentum in other asset classes, excess relative returns for this asset class are positively skewed and load favorably on risk metrics. Returns correlate with term premium estimates and yield curve factors, but substantial variance remains unexplained and the alphas are meaningfully positive.An extensive literature documents a sizable momentum anomaly across financial asset classes. Portfolios of securities with the most positive (negative) prior excess returns subsequently tend to have superior (inferior) risk-adjusted results. Studies have documented momentum profits with respect to individual shares, aggregate equity market indexes, currencies, commodities, and speculative-grade, rather than investment-grade, corporate bonds. Empirical tests usually consist of sorting portfolios on the basis of past returns, going long or overweighting past winners while shorting or underweighting past losers, calculating corresponding returns, and estimating risk exposure. Despite strong incentives, to date, analyses of cross-sectional momentum along government bond term structures (i.e., momentum patterns with respect to duration buckets along the curve), rather than across different markets or individual curves over time, remain unreported. This study finds sizable excess returns, notably under the index-duration-neutral constraint, of up to 57 bps in annual terms, with information ratios (IRs) as great as 0.66, given all necessary available US Treasury data to cover a breadth of maturities of up to 30 years from March 1985 through December 2013. A strategy based on a weighted average of signals across 20 alternative look-back windows produces average excess returns (over the index) of about 34 bps and an IR of about 0.46. Unlike momentum strategies in other asset classes or carry trades, relative returns are skewed to the upside and do not load on common risk factors. If anything, they correlate favorably as a hedge against risky assets and liquidity proxies. Moreover, in contrast to a common misconception, the short side does not predominantly drive returns.Perhaps unfortunately for practitioners, these momentum returns are not conditioned on ex ante factors. There is limited evidence that term-structure momentum is “Treasury market state dependent,” at least contemporaneously, given a positive correlation between overall market and momentum returns. But the result is much less pronounced compared with evidence on shares, and contemporaneous conditionality does not connote risk exposure per se. Also, there is no positive relationship between the underlying magnitude of the momentum signal and subsequent returns. In addition, on the basis of the hypothesis that momentum may reflect compensation consistent with forward term premium estimates, I found that momentum returns do correlate to a degree with returns from portfolios based on arbitrage-free Gaussian affine term-structure models. Excess relative returns correlate with some principal components of the yield curve as well as key lagged forward rates. Nonetheless, substantial variance remains unexplained, the betas are less than 1, and the alphas are largely positive as well as demonstrably more robust compared with other published fixed-income momentum anomalies. Lastly, these results do not constitute confirmation of previously published findings of duration-based momentum in bond markets; rather, they represent unreported return persistence unrelated to the level of the term structure.Editor’s note: This article was reviewed and accepted by Executive Editor Robert Litterman.