Paper: arXiv 2510.23150

Authors: Alban Etienne, Jean-Jacques Ohana, Eric Benhamou, Béatrice Guez, Ethan Setrouk, Thomas Jacquot

Abstract

Recent work has emphasized the diversification benefits of combining trend signals across multiple horizons, with the medium-term window-typically six months to one year-long viewed as the “sweet spot” of trend-following. This paper revisits this conventional view by reallocating exposure dynamically across horizons using a Bayesian optimization framework designed to learn the optimal weights assigned to each trend horizon at the asset level. The common practice of equal weighting implicitly assumes that all assets benefit equally from all horizons; we show that this assumption is both theoretically and empirically suboptimal. We first optimize the horizon-level weights at the asset level to maximize the informativeness of trend signals before applying Bayesian graphical models-with sparsity and turnover control-to allocate dynamically across assets. The key finding is that the medium-term band contributes little incremental performance or diversification once short- and long-term components are included. Removing the 125-day layer improves Sharpe ratios and drawdown efficiency while maintaining benchmark correlation. We then rationalize this outcome through a minimum-variance formulation, showing that the medium-term horizon largely overlaps with its neighboring horizons. The resulting “barbell” structure-combining short- and long-term trends-captures most of the performance while reducing model complexity. This result challenges the common belief that more horizons always improve diversification and suggests that some forms of time-scale diversification may conceal unnecessary redundancy in trend premia.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced statistical methods including Bayesian optimization, graphical models with sparsity constraints, and theoretical derivations of minimum-variance allocation, but is also heavily data-driven with a comprehensive out-of-sample backtest framework across multiple assets, cost modeling, and ablation studies.

Research Flowchart

  flowchart TD
  RQ["Research Question<br>Revisit Multi-Horizon Trend Diversification<br>Is medium-term always optimal?"] --> Input["Data & Inputs<br>Asset-level trend signals<br>across short, medium, long horizons"]
  Input --> AO["Key Method 1: Asset-Level Optimization<br>Bayesian optimization learns optimal<br>horizon weights per asset"]
  AO --> BG["Key Method 2: Dynamic Allocation<br>Bayesian graphical model with<br>sparsity & turnover control"]
  BG --> Find["Key Findings: Barbell Structure<br>Medium-term adds redundancy<br>Short + Long horizons dominate"]
  Find --> Out["Outcomes<br>Higher Sharpe, better drawdowns<br>Lower complexity, maintained correlation"]
  AO -.-> MinVar["Rationalization<br>Minimum-variance formulation shows<br>high overlap in medium-term"]
  MinVar -.-> Find

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a Bayesian optimization framework to dynamically allocate horizon weights at the asset level, leading to a barbell structure that combines short- and long-term trends while excluding the medium-term horizon.

Source sentence
We address this limitation by dynamically allocating the weights assigned to each trend horizon through Bayesian optimization.

Assumes 2

  • Equal weighting across horizons implicitly assumes that all assets benefit equally from all horizons. model
    Source sentence
    The conventional practice of assigning equal weights to trend horizons implicitly assumes that all assets benefit equally from each horizon.
  • Transaction, roll and management fees are applied as fixed basis‑point costs to the backtest. execution
    Source sentence
    For any backtest, we use three layers of implementation costs (see Table1 for contract‑level figures): • Transaction cost (Tx.Cost). … •

Data

Backtest sample: 2005 – 2025 · CTA futures · multi-asset · daily

  • Market futures Universe market-prices unknown
    Commodities, Equity, Fixed income, FX futures
    Source sentence
    Table 1: Market futures Universe with Transaction / Roll costs in basis points.

Costs

Modeled Costs are part of the reported headline results.

Tx cost 2/15 bps (Commodities, Equity), 2/10 bps (Fixed income), 2/2 bps (FX); Roll cost measured as 2000–2025 average front‑to‑next spread; Management fee 50 bps per annum.

Source sentence
For any backtest, we use three layers of implementation costs (see Table1 for contract‑level figures): • Transaction cost (Tx.Cost). … •

Finds 5

Reported performance (No 125 (medium-term horizon excluded) CTA allocation): Sharpe 0.77

  • Excluding the 125‑day medium‑term horizon yields the highest overall Z‑score (+0.80) with uniformly positive contributions (+0.84 Sharpe, +0.96 Return/Max‑DD, +0.59 correlation). performance
    Source sentence
    As shown in Table 7, excluding the 125‑day horizon (No 125) yields the highest overall Z‑score (+0.80), with uniformly positive contributions across all three metrics: +0.84 for Sharpe, +0.96 for return over maximum drawdown ratio, and +0.59 for correlation.
  • The medium‑term horizon contributes little incremental performance or diversification once short‑ and long‑term components are included. theoretical
    Source sentence
    The key finding is that the medium‑term band contributes little incremental performance or diversification once short‑ and long‑term components are included.
  • Removing the 125‑day sleeve improves Sharpe ratios across subperiods, e.g., 0.77 vs. 0.74 full‑sample average and 1.41 vs. 1.37 in 2010‑2015. performance
    Source sentence
    Its Sharpe ratio exceeds that of the baseline “All Horizons” allocation in three of the four subperiods—2010–2015 (1.41 vs. 1.37), 2020–2025 (0.44 vs. 0.35), and in the full‑sample average (0.77 vs. 0.74).
  • Excluding the long‑term 500‑day horizon deteriorates performance (overall Z‑score –1.12, Sharpe 0.67 vs. 0.74 baseline). negative
    Source sentence
    In contrast, removing the 500‑day horizon (No 500) produces the worst outcome (–1.12), driven by simultaneous deterioration in both risk‑adjusted return (–0.86)
  • The resulting “barbell” structure—combining short‑ and long‑term trends—captures most performance while reducing model complexity. theoretical
    Source sentence
    The resulting "barbell" structure—combining short‑ and long‑term trends—captures most of the performance while reducing model complexity.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 24 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (25)

  1. Re-evaluating Short- and Long-Term Trend Factors in CTA Replication: A Bayesian Graphical Approach — → 2507.15876
  2. Trend Following: Is This Time Different? (2025) — link
  3. Time Series Momentum Factors (2025) — link
  4. Trend Following: Timing Fast And Slow Trends (2022) — link
  5. A Century Of Evidence On Trend-Following Investing (2017) — link
  6. Don’t Mistake Style for Skill: The Impact of Style Factors on Performance Among Trend Followers (2017)
  7. Time Series Momentum (2012) — link
  8. The Need For Speed In Trend Following Strategies (2023) — link
and 17 more
  1. Momentum Turning Points (2023) — link
  2. Has Trend Gone Flat? Return Convexity in Trend Following (2021)
  3. Optuna: A Next-generation Hyperparameter Optimization Framework (2019) — link
  4. Historical Performance of Trend Following (2013) — link
  5. The risk in hedge fund strategies: Theory and evidence from trend followers (2001)
  6. Dissecting Investment Strategies in the Cross‑Section and Time Series (2015) — link
  7. Momentum Strategies in Futures Markets and Trend‑Following Funds (2013) — link
  8. Two Centuries of Trend Following (2014)
  9. Portfolio Selection (1952) — link
  10. Modern Portfolio Theory and Investment Analysis (2014)
  11. Convex Optimization (2004) — link
  12. Trend Following Strategies: A Practical Guide (2025) — link
  13. The Markowitz Optimization Enigma: Is ‘Optimized’ Optimal? (1989) — link
  14. Optimal Versus Naive Diversification: How Inefficient is the 1/N Portfolio Strategy? (2009) — link
  15. Deep Decoding of Strategies (2022) — link
  16. Generative AI: Crafting Portfolios Tailored to Investor Preferences (2024) — link
  17. Momentum Trading: ‘Skews Me’ (2012)

1 of 25 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.