Regime-Based Momentum in Rates

[WITH CODE] Achieving a high-Sharpe ratio momentum strategy in the rates market

Hello!

Today, we will be investigating momentum in the U.S. Treasuries market.

First, I will highlight two academic papers that influenced this post. Then, I will showcase the analysis and a preliminary strategy with a Sharpe ratio of 1.2. This will likely turn into a series, as I am confident that these ideas can be improved.

Let’s get into it.

---

Academic Literature

The papers below find significant evidence of momentum in the U.S. Treasury market, identifying persistent return patterns that are not fully explained by standard financial models.

Sihvonen’s¹ research focuses on time-series momentum, where a bond’s own past returns predict its future returns. He finds this effect is driven primarily by autocorrelation in yield changes, rather than by bond “carry”.

Figure 2: shows the mean returns for different maturity bonds both for the full sample and in subsamples following positive and negative past month returns.

Figure 3: shows the annualized Sharpe ratios for different maturity bonds both for the full sample and in subsamples following positive and negative past month returns.

Because yield changes are tied to monetary policy, this phenomenon is also related to the post-FOMC announcement drift (which Alpha in Academia has explored in prior paid-only posts). The paper’s central conclusion is that this momentum factor is “unspanned” (which simply means that the momentum factor provides new, additional predictive information about future bond returns that is not already contained in the current yield curve). This finding contradicts standard term-structure models, which imply all predictive information should already be priced into current yields.

Durham’s paper², in contrast, investigates cross-sectional momentum by creating trading rules that compare different “duration buckets” along the yield curve. He designs a strategy that goes long the bonds that were “winners” and short the “losers,” while remaining “duration-neutral” to insulate the portfolio from parallel shifts in the yield curve. This strategy generated “sizeable excess returns”; a long-short, zero-duration version yielded up to 207 basis points annually with a high information ratio.

Similar to Sihvonen’s findings, Durham concludes that these abnormal returns cannot be fully explained by standard Gaussian arbitrage-free affine term structure models (GATSMs). The momentum strategy demonstrated a “meaningfully positive” alpha, suggesting a persistent anomaly that is not simply compensation for known duration risks.

These papers prove that there is clear opportunity in this space, especially since Sihvonen’s paper was published in 2022. Now, let’s replicate and improve on these results.