Paper: arXiv 2308.12212
Authors: Xingyue Pu, Stefan Zohren, Stephen Roberts, Xiaowen Dong
Abstract
Network momentum provides a novel type of risk premium, which exploits the interconnections among assets in a financial network to predict future returns. However, the current process of constructing financial networks relies heavily on expensive databases and financial expertise, limiting accessibility for small-sized and academic institutions. Furthermore, the traditional approach treats network construction and portfolio optimisation as separate tasks, potentially hindering optimal portfolio performance. To address these challenges, we propose L2GMOM, an end-to-end machine learning framework that simultaneously learns financial networks and optimises trading signals for network momentum strategies. The model of L2GMOM is a neural network with a highly interpretable forward propagation architecture, which is derived from algorithm unrolling. The L2GMOM is flexible and can be trained with diverse loss functions for portfolio performance, e.g. the negative Sharpe ratio. Backtesting on 64 continuous future contracts demonstrates a significant improvement in portfolio profitability and risk control, with a Sharpe ratio of 1.74 across a 20-year period.
Complexity vs Empirical Score
- Math Complexity: 8.0/10
- Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs only partially modelled; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.0)
Why this score: The paper presents a novel neural network architecture derived from algorithm unrolling, involves complex optimization and graph theory, and shows strong empirical results with a 20-year backtest on 64 futures contracts, achieving a Sharpe ratio of 1.74.
Research Flowchart
flowchart TD
Start["Research Goal:<br>How to improve Network Momentum<br>optimization end-to-end?"] --> Input["Input Data:<br>64 Continuous Future Contracts<br>(20-year Backtest)"]
Input --> Method["Proposed Method:<br>L2GMOM Framework<br>End-to-End ML (Algorithm Unrolling)"]
subgraph Process ["Computational Process"]
direction TB
Method --> Learn["Simultaneous Learning:<br>Financial Network +<br>Portfolio Signals"]
Learn --> Loss["Optimization:<br>Loss Function (e.g.,<br>Negative Sharpe Ratio)"]
end
Process --> Outcomes["Key Findings & Outcomes:<br>Improved Profitability & Risk Control<br>Sharpe Ratio: 1.74<br>High Interpretability"]
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 L2GMOM, an end-to-end machine learning framework that simultaneously learns financial networks and optimises trading signals for network momentum strategies.
Source sentence
To address these challenges, we propose L2GMOM, an end-to-end machine learning framework that simultaneously learns financial networks and optimises trading signals for network momentum strategies.
Assumes 5
- Momentum spillover is subtly echoed in the historical pricing data. data
Source sentence
adopt the reasonable assumption that momentum spillover is subtly echoed in the historical pricing data, which are more readily accessible.
- An asset’s future performance can be influenced by the past performance of its linked assets. market
Source sentence
Network momentum strategy takes into account the interconnections of assets, assuming an asset’s future performance can be influenced by the past performance of its linked assets.
- Graph adjacency matrices have non‑negative edges. model
Source sentence
from Eq.(10b) contains only non-negative edges, as a result of enforcing the constraints of Eq.(9).
- Volatility scaling targets a 15% annualised portfolio volatility. execution
Source sentence
we apply an additional layer of volatility scaling at the portfolio level to an annualised volatility target of 15% and report the above evaluation metrics.
Show 1 more
- Turnover costs are measured in basis points per trade. execution
Source sentence
is the turnover cost in bps, varying from 0.5 to 5 bps.
Data
future contracts · futures
- 64 ratio‑adjusted continuous future contracts market-prices commercial
Source sentence
Our dataset contains 64 ratio‑adjusted continuous future contracts obtained from the Pinnacle Data Crop CLC Database.
Costs
Unclear - sources disagree
Source sentence (the authors' words)
is the turnover cost in bps, varying from 0.5 to 5 bps. As ex‑ante costs rise, all strategies see a drop in cost‑adjusted Sharpe ratios.
Finds 4
Sharpe 1.74
- L2GMOM achieves a Sharpe ratio of 1.74 over a 20‑year backtest on 64 futures contracts. performance
Source sentence
Backtesting on 64 continuous future contracts demonstrates a significant improvement in portfolio profitability and risk control, with a Sharpe ratio of 1.74 across a 20-year period.
- L2GMOM and L2GMOM_SR outperform benchmark strategies in both profitability and risk metrics. performance
Source sentence
... L2GMOM also outperform‑ing GLinReg indicates that optimising the graph with respect to portfolio performance can yield higher returns.
- L2GMOM_SR attains the lowest volatility, downside deviation, maximum drawdown and MDD duration among the tested models. performance
Source sentence
L2GMOM_SR has the lowest volatility, downside deviation, MDD, and MDD duration in the raw signal, indicating that optimising with the Sharpe ratio can significantly reduce risk.
- Graph‑learning strategies (GLinReg, L2GMOM, L2GMOM_SR) show better resilience to increasing turnover costs than non‑graph models. performance
Source sentence
The proposed models, L2GMOM and L2GMOM_SR, also stay positive at 3 bps, showing resilience to increasing costs. Thus, graph learning strategies like GLinReg, L2GMOM, and L2GMOM_SR offer better cost resilience than LinReg.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 27 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.