Paper: arXiv 2501.07135
Authors: Linze Li, William Ferreira
Abstract
We present a systematic, trend-following strategy, applied to commodity futures markets, that combines univariate trend indicators with cross-sectional trend indicators that capture so-called {"\em momentum spillover"}, which can occur when there is a lead-lag relationship between the trending behaviour of different markets. Our strategy utilises two methods for detecting lead-lag relationships, with a method for computing {"\em network momentum"}, to produce a novel trend-following indicator. We use our new trend indicator to construct a portfolio whose performance we compare to a baseline model which uses only univariate indicators, and demonstrate statistically significant improvements in Sharpe ratio, skewness of returns, and downside performance, using synthetic bootstrapped data samples taken from time-series of actual prices.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 6.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced mathematics, including Lévy areas (stochastic calculus), dynamic time warping, graph theory (adjacency matrices), and convex optimization, indicating high complexity. Empirical rigor is solid with backtests on commodity futures, bootstrap sampling, and statistical significance testing (Sharpe ratios), though it lacks raw code or high-frequency data implementation.
Research Flowchart
flowchart TD A["Research Goal"] -->|"Identify"| B["Input Data"] B -->|"Compute"| C["Univariate Trends"] B -->|"Detect"| D["Lead-Lag Relationships"] C & D -->|"Compute"| E["Network Momentum"] E -->|"Construct"| F["Novel Trend Indicator"] F -->|"Form Portfolio"| G["Performance Analysis"] G -->|"Compare to"| H["Baseline Model"] H -->|"Outcome"| I["Key Findings"] A["Research Goal<br>Enhance trend-following<br>using momentum spillover"] B["Input Data<br>Commodity futures price series"] C["Univariate Trends<br>Market-specific indicators"] D["Lead-Lag Detection<br>Two methods used"] E["Network Momentum<br>Cross-sectional computation"] F["Novel Trend Indicator<br>Combined approach"] G["Performance Analysis<br>Portfolio construction"] H["Baseline Model<br>Univariate only"] I["Key Findings<br>Improved Sharpe ratio<br>Better skewness<br>Superior downside performance"]
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 methodology that transforms cross‑sectional momentum spillover into network momentum and uses it to build a systematic trend‑following portfolio.
Source sentence
We propose a methodology that transforms cross-sectional momentum spillover into network momentum across market industries.
Assumes 5
- The functional form of the relationship between standardized returns is given by ˜r_n,t = βℓ f(˜r_m,t-ℓ) + ε_t. model
Source sentence
suppose the functional form of the relationship between the standardised returns on asset m and asset n is given by: ˜rn,t = βℓf(˜rm,t−ℓ) + ϵt,
- P&L from a signal at time t is realised on the position at time t+2, reflecting a one‑day delay in trade execution. execution
Source sentence
We take a conservative approach to pnl calculation, assuming that a signal/position generated at time t, including information up to time t, is established via a trade at time t + 1, and then pnl is earned on the position at time t + 2.
- Transaction costs are approximated by half the average bid‑ask spread for each market. market
Source sentence
We choose market specific transaction costs based on half the average bid-ask spread in the market, denoted sm, over some representative historical period.
- Stationary block bootstrap preserves the auto‑ and cross‑correlation structure of market returns. statistical
Source sentence
Stationary block bootstrapping preserves the auto- and cross-correlation of market returns, and so enables us to make statistical comparisons about the relative performance of the different models.
Show 1 more
- The lead‑lag matrix obtained via dynamic time warping contains only integer lags. model
Source sentence
First, the lead-lag matrix, obtained via dynamic time warping, contains only integer lags, i.e., Vi,j = ℓi,j ∈Z, which indicates that ˜ri,t−ℓi,j has some predictive power over ˜rj,t, but not its magnitude.
Data
2005 – 2024 · commodity futures · futures · daily
- 28 commodity futures contracts market-prices commercial
Source sentence
price of 28 futures markets, ranging across sectors: agriculture, energy, metals and equity indices. The model training period spans from June 2002 to June 2024, with strategy performance evaluated on out-of-sample data from January 2005 to June 2024.
- Bootstrapped price trajectories synthetic unknown
Source sentence
... to generate a population set of price trajectories using the true market data.
Costs
Modeled
Costs are set to half the average bid‑ask spread for each market (sm) and applied to position changes.
Source sentence
We choose market specific transaction costs based on half the average bid-ask spread in the market, denoted sm, over some representative historical period.
Finds 5
Sharpe 0.36 (benchmark 0.23)
- Network momentum models achieve higher net Sharpe ratios than the baseline MACD model, with NMM‑DDTW‑E reaching a Sharpe of 0.357 (29% improvement) on bootstrapped data and 0.364 on real data. performance
Source sentence
... NMM‑DTW‑E achieves the highest net Sharpe ratio at 0.364, compared to the benchmark’s 0.233.
- Ensemble network momentum models reduce transaction costs relative to the baseline, e.g., DTW variations lower costs to 0.022, about 19% lower than MACD. performance
Source sentence
However, ensemble methods reduce transaction costs, with DTW variations further decreasing them to 0.022, approximately 19% lower than MACD.
- NMM‑LEVY shows the smallest maximum drawdown and highest hit rate, indicating effective trend reversal identification. performance
Source sentence
NMM‑LEVY demonstrates the smallest MDD and highest hit rate, suggesting it is particularly effective at identifying trend reversals.
- Network momentum signals are low‑correlated with MACD (average correlation 0.71‑0.89), providing diversification benefits. performance
Source sentence
By analysing the returns between the NMM models and the benchmark MACD
- The portfolio constructed with network momentum yields more positively skewed returns than the baseline. performance
Source sentence
The proposed framework also consistently yields more positively skewed returns, underscoring the efficiency and robustness of the network momentum identified for trend‑following strategies.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 34 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 (57)
- The Stationary Bootstrap (1994) — link
- Dissecting Investment Strategies in the Cross Section and Time Series (2015) — link
- Building Cross-Sectional Systematic Strategies By Learning to Rank (2021) — link
- Design and analysis of momentum trading strategies (2023) — link
- Profitability of Momentum Strategies: An Evaluation of Alternative Explanations (1999)
- Dynamic Time Warping for Lead-Lag Relationships in Lagged Multi-Factor Models — → 2309.08800
- Derivative dynamic time warping (2001)
- Lead-Lag Estimation by Means of the Dynamic Time Warping Technique (2001) — link
and 49 more
- shapeDTW: Shape Dynamic Time Warping (2018) — link
- Detecting Lead-Lag Relationships in Stock Returns and Portfolio Strategies (2023)
- Lead-lag detection and network clustering for multivariate time series with an application to the US equity market (2022) — link
- A Primer on the Signature Method in Machine Learning (2016) — link
- Dynamic time warping (2007) — link
- Addressing big data time series: Mining trillions of time series subsequences under dynamic time warping (2013)
- CVXPY: A Python-Embedded Modeling Language for Convex Optimization (2016) — link
- How to learn a graph from smooth signals (2016)
- Individual comparisons by ranking methods (1992)
- Generalizing DTW to the multi-dimensional case requires an adaptive approach (2017)
- Extracting information from the signature of a financial data stream (2014) — link
- Continuous Futures Contracts Methodology for Backtesting (2019) — link
- Network Momentum across Asset Classes — → 2308.11294
- Enhancing Time Series Momentum Strategies Using Deep Neural Networks (2020) — link
- Momentum Profits, Factor Pricing, and Macroeconomic Risk (2008) — link
- A Unified Theory of Underreaction, Momentum Trading and Overreaction in Asset Markets (1999)
- Industry information diffusion and the lead-lag effect in stock returns (2007)
- A model of investor sentiment (1998)
- An institutional theory of momentum and reversal (2013)
- Returns to buying winners and selling losers: Implications for stock market efficiency (1993)
- International momentum strategies (1998)
- Momentum, ownership structure, and financial crises: An analysis of Asian stock markets (2000)
- Profitability of momentum strategies: An evaluation of alternative explanations (2001)
- A century of evidence on trend-following investing (2017)
- Stock and bond market interaction: Does momentum spill over? (2005)
- When are Contrarian Profits Due to Stock Market Overreaction? (1990)
- Of Shepherds, Sheep, and the Cross-autocorrelations in Equity Returns (1995)
- Investment analysis and the adjustment of stock prices to common information (1993)
- The lead–lag relationship between the spot and futures markets in China (2017)
- Detecting leaders from correlated time series (2010)
- Econometric measures of connectedness and systemic risk in the finance and insurance sectors (2012)
- The econometrics of financial markets (1998)
- A new measure of rank correlation (1938)
- Measuring and testing dependence by correlation of distances (2007)
- Information-theoretic approach to lead-lag effect on financial markets (2014)
- Exploring and measuring non-linear correlations: Copulas, Lightspeed Transportation and Clustering (2017)
- Do industries explain momentum? (1999)
- Risk-managed industry momentum and momentum crashes (2018)
- Shared analyst coverage: Unifying momentum spillover effects (2020)
- Spatio-temporal momentum: Jointly learning time-series and cross-sectional strategies — → 2302.10175
- Building cross-sectional systematic strategies by learning to rank (2020) — link
- Lead–lag relationships in foreign exchange markets (2020)
- Momentum information propagation through global supply chain networks (2021)
- Momentum spillover from stocks to corporate bonds (2017)
- Trend-Following and Spillover Effects (2019)
- Currency News and International Bond Markets (2021)
- Cross-asset time-series momentum: Crude oil volatility and global stock markets (2023)
- Kolmogorov–smirnov test: Overview (2014)
- Speech discrimination by dynamic programming (1968)
3 of 57 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.