Skip to content

Optimal Allocation to Time-Series and Cross-Sectional Momentum: summary and key takeaways

The paper derives how much risk a trend follower should put in absolute (time-series) versus relative (cross-sectional) trends. The answer depends on trend strengths and correlations. In a 59-futures simulation, the resulting dynamic strategy had a higher Sharpe ratio than either pure strategy, with an average split near 80% absolute and 20% relative.

Schmid, O.; Wirth, P. · 2021 · The Journal of Portfolio ManagementSummary updated 20 November 2026
Published2021Journal article
Asset focusMulti-assetMomentum
TypeBothTrend & Momentum
Reading time4 minthis summary

Optimal Allocation to Time-Series and Cross-Sectional Momentum at a glance

The question
What is the ex ante optimal mix of time-series and cross-sectional momentum in a pure trend-following strategy, and does it pay off out of the theory?
Dataset & period
Daily settlement prices of 59 liquid futures from Bloomberg (equity indexes, commodities, government bonds, short-term interest rates, currency pairs against the USD). Abstract and exhibits state 2000 to 2018; the data section says from January 2001 through December 2018. Daily frequency.
Method
Each pair of risk-adjusted instruments is treated as a small optimisation: maximise expected trend strength for a fixed risk budget. The closed-form solution depends only on the two trend signals and their correlation. Low correlation pushes the weights toward pure time-series positions. High correlation pushes them toward a long-short pair trade. Signals are moving-average crossovers scaled by volatility, mapped through a bounded logistic function, and built bottom-up across all pairs. Three strategies (dynamic, absolute, relative) plus a 50/50 mix are simulated with portfolios formed on prior-day data and traded one day later.

Key findings of Optimal Allocation to Time-Series and Cross-Sectional Momentum

  • The paper's average ex ante optimal split is around 80% time-series and 20% cross-sectional momentum, which fits earlier work saying time-series momentum is generally stronger.
  • At the middle signal setting without costs, the Sharpe ratios are 0.88 (dynamic), 0.69 (absolute) and 0.53 (relative). With costs they are 0.70, 0.48 and 0.35.
  • The dynamic strategy's Sharpe advantage over each pure strategy and over the 50/50 mix is significant in the paper's Memmel tests, for every signal setting, with and without costs.
  • A fixed 50/50 mix lands between the two pure strategies on Sharpe ratio, so a naive blend does not add value. The paper links this to the very high correlation among the strategies.
  • With costs at a 10% volatility target, the dynamic strategy's maximum loss is 18.6%, against 24.9% and 24.6% for the pure strategies. It has slightly lower skewness.
  • The relative-only carve-out (absolute component set to zero) has the highest Sharpe ratio before costs and positive daily skewness, but very high turnover. With costs its Sharpe ratio falls to 0.56 at the middle setting, below the dynamic strategy.

Robustness, as reported in the paper

  • Costs are tested: the simulation is run with and without a per-lot cost of $20, which the authors call conservative. The dynamic strategy stays the best of the three main strategies after costs.
  • The dynamic strategy turns over less than or equal to the pure strategies, so it loses less Sharpe ratio to costs.
  • Average allocation is reported for sub-periods. The relative share is somewhat higher in the post-2008 crisis period, consistent with higher correlations then.
  • Results are shown for three moving-average parametrizations. The paper says the choice of logistic versus tanh mapping does not matter much, and that crisis behaviour is similar across signal settings.

Caveats and limits

  • The sample is short and starts in 2000 or 2001 by design, to keep a broad futures universe. The paper notes earlier data would shrink the universe and tilt it to commodities.
  • Costs are a flat per-lot assumption that also covers rolls and slippage. After costs, the relative strategy's mean return is not statistically different from zero at the fastest setting (inferred from the t-values).
  • The relative carve-out looks strong before costs, but the paper itself says its turnover is three to four times higher than other strategies' and that turnover reduction is unsolved.
  • The paper's own authors note that results for the absolute carve-out are not shown, and the studies it cites are exposed to data snooping.

Items marked (inferred) are our reading of the design, not statements by the authors.

Why it matters

A CTA or trend-following book usually picks time-series or cross-sectional signals once and keeps the mix fixed. This paper gives a rule for varying that mix with correlation and trend dispersion. It also shows why a static blend of two highly correlated trend sleeves adds little.

Questions about this paper

When does time-series momentum beat cross-sectional momentum?

The paper says time-series momentum does better when instruments share similar trends and correlations are low. Cross-sectional momentum is preferable when correlations are high and trend dispersion is large.

What does the optimal mix look like on average?

Over the futures sample, the average ex ante optimal allocation is around 80% to time-series and 20% to cross-sectional momentum.

Does a simple 50/50 blend of the two work as well?

No. In the paper the constant 50/50 mix has a Sharpe ratio between those of the two pure strategies, while the dynamic allocation beats both.

Where can I read the original paper?

The original is at its publisher: https://doi.org/10.3905/jpm.2021.1.213. We link to the paper and do not host the PDF.

Follow new quant research

Download MacroFinance market data as CSV or Excel, or pull it by API, and test the ideas in these papers yourself. New summaries are added as we review them.

Create a free account
Summaries are written by MacroFinance from the published papers, in our own words, for information only. They are not investment advice and may omit detail. Papers are not hosted here: each page links to the original at its publisher.