Paper: arXiv 2308.11294

Authors: Xingyue Pu, Stephen Roberts, Xiaowen Dong, Stefan Zohren

Abstract

We investigate the concept of network momentum, a novel trading signal derived from momentum spillover across assets. Initially observed within the confines of pairwise economic and fundamental ties, such as the stock-bond connection of the same company and stocks linked through supply-demand chains, momentum spillover implies a propagation of momentum risk premium from one asset to another. The similarity of momentum risk premium, exemplified by co-movement patterns, has been spotted across multiple asset classes including commodities, equities, bonds and currencies. However, studying the network effect of momentum spillover across these classes has been challenging due to a lack of readily available common characteristics or economic ties beyond the company level. In this paper, we explore the interconnections of momentum features across a diverse range of 64 continuous future contracts spanning these four classes. We utilise a linear and interpretable graph learning model with minimal assumptions to reveal the intricacies of the momentum spillover network. By leveraging the learned networks, we construct a network momentum strategy that exhibits a Sharpe ratio of 1.5 and an annual return of 22%, after volatility scaling, from 2000 to 2022. This paper pioneers the examination of momentum spillover across multiple asset classes using only pricing data, presents a multi-asset investment strategy based on network momentum, and underscores the effectiveness of this strategy through robust empirical analysis.

Complexity vs Empirical Score

  • Math Complexity: 6.0/10
  • Empirical Rigor: 6.0/10 (capped from 8.5: transaction costs only partially modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 8.5)

Why this score: The paper employs a linear graph learning model, a mathematically advanced concept requiring matrix operations and optimization, but keeps it interpretable; it demonstrates high empirical rigor with a multi-asset backtest over 22 years (2000-2022), robustness analysis, and specific performance metrics like Sharpe ratio and annual returns.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Investigate Network Momentum<br>across 64 Futures"] --> B["Data Input<br>64 Continuous Futures<br>Commodities, Equities, Bonds, Currencies"]
  B --> C["Methodology<br>Linear Graph Learning Model"]
  C --> D["Computation<br>Construct Momentum<br>Spillover Network"]
  D --> E["Strategy<br>Network Momentum Strategy<br>Volatility Scaling"]
  E --> F["Outcomes<br>Sharpe Ratio: 1.5<br>Annual Return: 22%<br>2000-2022"]

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 graph learning model to infer dynamic networks among futures contracts and builds a network momentum trading strategy.

Source sentence
we propose a novel approach using a graph learning model [26] to infer dynamic networks among 64 continuous futures contracts across four asset classes

Assumes 5

  • Graph signals are low‑frequency (low‑pass) components in the graph spectral domain. model
    Source sentence
    graph learning involves estimating a graph adjacency or Laplacian matrix with a model assumption that graph signals mainly consists of low frequency components in the graph spectral domain, i.e. low-pass graph signals
  • Each asset’s momentum features are linearly combined with edge weights to form network momentum covariates. model
    Source sentence
    For each asset, the covariates are network momentum features, which are weighted average of its connected assets’ individual momentum features, with edge values as weights
  • The linear regression model predicts a 1‑day volatility‑scaled return. model
    Source sentence
    The model aims to predict an asset’s future trend, targeted by its future 1‑day volatility‑scaled return
  • Turnover regularisation is not applied in the current strategy. execution
    Source sentence
    We reserve this for future research
Show 1 more
  • All assets are treated as nodes in a symmetric, non‑negative adjacency matrix. model
    Source sentence
    The constraints guarantee the learned graph adjacency matrix is symmetric and non‑negative

Data

Backtest sample: 2000 – 2022 · futures · multi-asset

  • Daily prices of 64 continuous futures contracts · Pinnacle Data Crop CLC Database market-prices commercial
    1990‑2022 (varying per asset) · Commodities, Equities, Fixed Income, Currencies futures
    Source sentence
    Our raw dataset contains the daily prices of 64 highly liquid ratio‑adjusted continuous futures contracts. These contracts are extracted from the Pinnacle Data Crop CLC Database2. The

Costs

Unclear - sources disagree The platform classifier and the paper-text extraction give different answers, so no value is asserted.

Platform classifier: Partial

Paper text extraction: Modeled - Cost‑adjusted Sharpe is computed for pseudo‑costs c = {0, 0.5, 1, 2, 3, 4, 5} basis points.

Source sentence (the authors' words)
c is the pseudo‑cost in basis points (bps). We consider c = {0, 0.5, 1, 2, 3, 4, 5} bps in Figure 3a

Finds 4

Reported performance (network momentum strategy): Sharpe 1.50 · annual return 22.0%

  • The network momentum strategy achieves a Sharpe ratio of 1.5 and an annual return of 22% after volatility scaling (2000‑2022). performance
    Source sentence
    By leveraging the learned networks, we construct a network momentum strategy that exhibits a Sharpe ratio of 1.5 and an annual return of 22%, after volatility scaling, from 2000 to 2022
  • GMOM maintains a positive Sharpe up to 3 bps transaction cost and above 0.5 up to 2 bps, outperforming LinReg under costs. performance
    Source sentence
    GMOM demonstrates resilience to transaction costs. It manages to maintain a positive Sharpe ratio up to a cost of 3bps, and a ratio above 0.5 up to 2bps
  • Model‑based methods (LinReg, GMOM, RegCombo, SignCombo) exhibit higher turnover than model‑free methods (MACD, Long Only). performance
    Source sentence
    model‑based methods, such as LinReg, GMOM, RegCombo, and SignCombo, exhibit higher turnovers, which is consistent with findings in the literature
  • Combining individual momentum and network momentum features (RegCombo) improves portfolio performance relative to using either alone. performance
    Source sentence
    a better portfolio can be generated when these two types of features are combined

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 16 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 6 papers in the archive (5 by independent authors)

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.