Paper: arXiv 2410.20597

Authors: Dragos Gorduza, Yaxuan Kong, Xiaowen Dong, Stefan Zohren

Abstract

We investigate the effectiveness of a momentum trading signal based on the coverage network of financial analysts. This signal builds on the key information-brokerage role financial sell-side analysts play in modern stock markets. The baskets of stocks covered by each analyst can be used to construct a network between firms whose edge weights represent the number of analysts jointly covering both firms. Although the link between financial analysts coverage and co-movement of firms’ stock prices has been investigated in the literature, little effort has been made to systematically learn the most effective combination of signals from firms covered jointly by analysts in order to benefit from any spillover effect. To fill this gap, we build a trading strategy which leverages the analyst coverage network using a graph attention network. More specifically, our model learns to aggregate information from individual firm features and signals from neighbouring firms in a node-level forecasting task. We develop a portfolio based on those predictions which we demonstrate to exhibit an annualized returns of 29.44% and a Sharpe ratio of 4.06 substantially outperforming market baselines and existing graph machine learning based frameworks. We further investigate the performance and robustness of this strategy through extensive empirical analysis. Our paper represents one of the first attempts in using graph machine learning to extract actionable knowledge from the analyst coverage network for practical financial applications.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs only partially modelled; reported Sharpe ratio of 4.069 is implausibly high for daily or lower-frequency data; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 8.0)

Why this score: The paper uses advanced graph machine learning (Graph Attention Networks) to model complex network relationships, indicating high math complexity. It also demonstrates high empirical rigor through extensive backtesting with specific performance metrics (29.44% returns, Sharpe 4.06), robustness checks, and benchmark comparisons using real financial data.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Extract Alpha from Analyst Networks"] --> B["Data Collection<br>Analyst Coverage & Stock Returns"]
  B --> C["Network Construction<br>Build Analyst Coverage Graph"]
  C --> D["Modeling<br>Graph Attention Network GNN"]
  D --> E["Forecasting<br>Node-Level Price Predictions"]
  E --> F["Portfolio Construction<br>Long/Short Strategy based on Signals"]
  F --> G["Key Findings<br>29.44% Ann. Return / 4.06 Sharpe Ratio"]

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 attention network–based trading strategy that leverages the analyst coverage network to generate stock return forecasts.

Source sentence
we build a trading strategy which leverages the analyst coverage network using a graph attention network.

Assumes 1

  • A monthly target frequency is appropriate for the effect of analyst coverage on investors to manifest. statistical
    Source sentence
    we assume the monthly frequency is a reasonable timeframe for the effect of analyst coverage on investors to manifest itself.

Data

Backtest sample: US stocks · equities · monthly

  • CRSP/COMPUSTAT stock price dataset · WRDS market-prices commercial
    2006-2022 · 495 firms of the S&P500
    Source sentence
    We investigate a dataset of stock prices extracted from the CRSPR/COMPUSTAT datasource hosted by WRDS covering the 2006-2022 period
  • IBES analyst estimates · IBES via WRDS fundamentals commercial
    2006-2022 · 495 firms of the S&P500
    Source sentence
    We combine this dataset with information from Institutional Brokers’ Estimate System (IBES) analyst estimates from 2006 to 2022.

Recognised sources: I/B/E/S Compustat WRDS

Costs

Partial Costs appear only as a robustness check, a flat fee without spread or impact, or outside the headline numbers.

Costs ranging from 0 to 5 basis points are applied to assess impact on Sharpe ratio and turnover.

Source sentence
annualized Sharpe ratio of the different strategies under increasing trading costs from 0 basis points increasing to 1, 2 and 5 basis points.

Finds 3

Reported performance (GAT analysts): Sharpe 4.07 (benchmark 0.41) · annual return 29.4% · max drawdown 6.0%

  • The proposed GAT‑analysts strategy achieves annualized returns of 29.44% with a Sharpe ratio of 4.06, substantially outperforming market baselines and existing graph‑ML frameworks. performance
    Source sentence
    We develop a portfolio based on those predictions which we demonstrate to exhibit an annualized returns of 29.44% and a Sharpe ratio of 4.06 substantially outperforming market baselines and existing graph machine learning based frameworks.
  • GAT‑analysts and GAT‑correlations maintain positive Sharpe ratios up to 2 basis points of transaction costs, whereas other models turn negative. performance
    Source sentence
    Only GATanalysts, GATcorr and GAT1_layer maintain positive Sharpe ratios when exposed to 2 basis points of trading costs: every other strategy exhibits a null or negative Sharpe ratio.
  • Ablation studies show that using the full analyst matrix improves cumulative log returns by 25% compared to random perturbations and by 63% compared to the analyst‑matrix‑only baseline. performance
    Source sentence
    having the complete analyst matrix instead of the randomly perturbed one leads to 63% more returns.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 21 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 (34)

  1. Shared analyst coverage: Unifying momentum spillover effects (2020)
  2. Modeling Momentum Spillover with Economic Links Discovered from Financial Documents (2023)
  3. Uncovering expected returns: Information in analyst coverage proxies (2017)
  4. The effect of innovation similarity on asset prices: Evidence from patents’ big data (2023)
  5. Information, analysts, and stock return comovement (2015)
  6. Industry information diffusion and the lead-lag effect in stock returns (2007)
  7. Do industries explain momentum? (1999)
  8. Complicated firms (2012)
and 26 more
  1. Text-based industry momentum (2018)
  2. Economic links and predictable returns (2008)
  3. Market segmentation and cross-predictability of returns (2010)
  4. Technological links and predictable returns (2019)
  5. Far away from home: Investors’ underreaction to geographically dispersed information (2022)
  6. Geographic lead-lag effects (2020)
  7. Institutional ownership and return predictability across economically unrelated stocks (2017)
  8. Alliances and return predictability (2016)
  9. Analyst coverage networks and corporate financial policies (2023)
  10. Common analyst links and predictable returns: Evidence from China (2022)
  11. Network momentum across asset classes — → 2308.11294
  12. Exploring graph neural networks for stock market predictions with rolling window analysis (2019) — link
  13. Multivariate realized volatility forecasting with graph neural network (2022)
  14. Shared Analyst Coverage and Connected-Firm Momentum Spillover in China (2023)
  15. The Analyst Matrix: Profiting from Sell-Side Analysts’ Coverage Networks (2020) — link
  16. Graph attention networks (2018)
  17. Dissecting investment strategies in the cross section and time series (2015)
  18. Observational Learning in Networks of Competition: How structures of attention among rivals can bring interpretive advantage (2023)
  19. Analyst coverage overlaps and interfirm information spillovers (2021)
  20. Does common analyst coverage explain excess comovement? (2016)
  21. Pair analyst coverage and return comovement: Evidence from China (2023)
  22. Network Momentum across Asset Classes — → 2308.11294
  23. Enhancing time series momentum strategies using deep neural networks (2019) — link
  24. Semi-supervised classification with graph convolutional networks (2016) — link
  25. Graph neural networks for asset management (2021)
  26. A review of two decades of correlations, hierarchies, networks and clustering in financial markets (2021)

2 of 34 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.