Paper: arXiv 2410.23447

Authors: Marcus Gawronsky, Chun-Sung Huang

Abstract

This paper introduces a novel approach to financial risk analysis that does not rely on traditional price and market data, instead using market news to model assets as distributions over a metric space of risk factors. By representing asset returns as integrals over the scalar field of these risk factors, we derive the covariance structure between asset returns. Utilizing encoder-only language models to embed this news data, we explore the relationships between asset return distributions through the concept of Energy Distance, establishing connections between distributional differences and excess returns co-movements. This data-agnostic approach provides new insights into portfolio diversification, risk management, and the construction of hedging strategies. Our findings have significant implications for both theoretical finance and practical risk management, offering a more robust framework for modelling complex financial systems without depending on conventional market data.

Complexity vs Empirical Score

  • Math Complexity: 8.0/10
  • Empirical Rigor: 2.5/10
  • Quadrant: Lab Rats — high math complexity, limited empirical testing

Why this score: The paper relies on advanced continuous mathematics, including integrals over risk factor spaces and rigorous statistical derivations, which is dense. However, it lacks empirical backtesting, specific datasets, or implementation code, focusing instead on theoretical modeling and NLP embeddings without reported results.

Research Flowchart

  flowchart TD
  A["Research Goal: Model Asset Correlations without Market Data"] --> B["Data Input: Raw Market News"]
  B --> C["Computation: Encoder-Only LLM Embeddings"]
  C --> D["Core Method: Energy Distance Metric"]
  D --> E["Analysis: Risk Factor Distributions"]
  E --> F["Outcome: Covariance & Excess Returns"]
  F --> G["Applications: Diversification & Hedging"]

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 novel, data‑agnostic risk analysis method that models assets as distributions over a continuous risk‑factor space using news‑derived embeddings and Energy Distance.

Source sentence
This paper introduces a novel approach to financial risk analysis that does not rely on traditional price and market data, instead using market news to model assets as distributions over a metric space of risk factors.

Assumes 3

  • Assets are represented as distributions over a continuous risk‑factor space rather than as point estimates. model
    Source sentence
    In this paper, we depart from the assumption in sAPT that exists as points in a scalar field of risk factors, and instead consider assets as distributions over this space.
  • Covariance between market risk premiums is mainly determined by diagonal terms; off‑diagonal contributions are captured by a residual term. statistical
    Source sentence
    under the assumption that the covariance between market risk premiums is predominantly determined by the diagonal terms and that the off-diagonal contributions are captured by ut
  • The residual term ut can be negligible or modeled as Gaussian noise depending on its stochastic properties. statistical
    Source sentence
    Depending on the characteristics of ϵt(ω, ω′), ut may be negligible or can be modeled as a Gaussian noise term if ϵt(ω, ω′) exhibits appropriate stochastic properties.

Data

  • News articles dataset text-news public
    2018-2022 · 53 Nasdaq‑listed companies
    Source sentence
    Our dataset comprises 66,000 news articles covering 53 companies listed on the Nasdaq, published between 2018 and 2022.

Costs

Not applicable The paper reports no trading or execution result.

Finds 2

  • The Mantel test shows a significant positive correlation (0.412, p‑value 0.0001) between Energy Distance of news‑derived embeddings and observed return correlations, supporting the Energy Distance inequality. statistical
    Source sentence
    The Mantel test yielded a Mantel correlation coefficient of 0.412 with a corresponding p-value of 0.0001. This significant result allows us to reject the null hypothesis that the Energy Distance between assets does not constrain the observed return correlations.
  • Energy Distance serves as an upper bound on asset return correlations, indicating that more similar latent risk‑factor distributions lead to higher return co‑movement. theoretical
    Source sentence
    This finding implies that the Energy Distance between assets serves as an upper bound on the correlations between their returns.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 17 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 (55)

  1. Decomposition and Characterization of Risk with a Continuum of Random Variables (1995) — link
  2. Distance-Based Tests for Homogeneity of Multivariate Dispersions (2006) — link
  3. A New Method for Non-Parametric Multivariate Analysis of Variance (2001) — link
  4. A Computational Fluid Mechanics Solution to the Monge-Kantorovich Mass Transfer Problem (2000) — link
  5. Spatial \ Dependence\ in \ Financial\ \ Data\ : \ Importance\ of the \ Weights\ \ Matrix\ (2016)
  6. Qualitative Similarity and Stock Price Comovement (2018) — link
  7. Meet the Press: Survey Evidence on Financial Journalists as Information Intermediaries (2022)
  8. Arbitrage, Factor Structure, and Mean-Variance Analysis on Large Asset Markets (1983)
and 47 more
  1. Expected Returns and Large Language Models (2026) — link
  2. Semiparametric Estimation of a Characteristic-Based Factor Model of Common Stock Returns (2007) — link
  3. Efficient Semiparametric Estimation of the Fama-French Model and Extensions (2012) — link
  4. Evidence on the Characteristics of Cross Sectional Variation in Stock Returns (1997)
  5. Cross-Section Without Factors: A String Model for Expected Returns (2024)
  6. Disclosure Similarity and Future Stock Return Comovement (2024) — link
  7. Multifactor Explanations of Asset Pricing Anomalies (1996) — link
  8. Spatial Linkages in International Financial Markets (2011) — link
  9. High-Dimensional Spatial Arbitrage Pricing Theory with Heterogeneous Interactions — → 2511.01271
  10. Should We Augment Large Covariance Matrix Estimation with Auxiliary Network Information? (2026)
  11. News-Implied Linkages and Local Dependency in the Equity Market (2023)
  12. Media Bias and Reputation (2006)
  13. On Stable Factor Structures in the Pricing of Risk: Do Time-Varying Betas Help or Hurt? (1998)
  14. Should We Still Pretrain Encoders with Masked Language Modeling? (2026) — link
  15. The Term Structure of Interest Rates as a Random Field (2000)
  16. Tobin’s Marginal q and Average q: A Neoclassical Interpretation (1982)
  17. Chronologically Consistent Large Language Models — → 2502.21206
  18. Scope, Scale, and Concentration: The 21st-Century Firm (2025) — link
  19. Text-Based Network Industries and Endogenous Product Differentiation (2016) — link
  20. \ FinBERT\ -/-/-A Large Language Model for Extracting Information from Financial Text (2023) — link
  21. On the Translocation of Masses (1958) — link
  22. Characteristics Are Covariances: A Unified Model of Risk and Return (2019)
  23. The Term Structure of Interest Rates as a Gaussian Random Field (1994)
  24. Asymptotic Arbitrage and the APT with or Without Measure-Theoretic Structures (2001) — link
  25. Asset Pricing with Spatial Interaction (2018) — link
  26. From Word Embeddings to Document Distances (2015) — link
  27. Matryoshka Representation Learning (2022) — link
  28. Can \ ChatGPT\ Forecast Stock Price Movements? Return Predictability and Large Language Models — → 2304.07619
  29. An Infinite-Dimensional Model of Liquidity in Financial Markets (2021)
  30. Market \ Segmentation\ and \ Cross\ -predictability of \ Returns\ (2010) — link
  31. Deriving the APT When the Number of Factors Is Unknown (2000) — link
  32. Nasdaq Listings Delivers Record-Breaking First Half (2026) — link
  33. An Invitation to Statistics in Wasserstein Space (2020) — link
  34. Permutation P-values Should Never Be Zero: Calculating Exact P-values When Permutations Are Randomly Drawn (2010)
  35. Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models (2025) — link
  36. Reference Variables, Factor Structure, and the Approximate Multibeta Representation (1992)
  37. Extra-Market Components of Covariance in Security Returns (1974) — link
  38. The Arbitrage Theory of Capital Asset Pricing (1976)
  39. The Dynamics of the Forward Interest Rate Curve with Stochastic String Shocks (2001) — link
  40. Optimal Transport for Applied Mathematicians: Calculus of Variations, PDEs, and Modeling (2015)
  41. Economic Linkages Inferred from News Stories and the Predictability of Stock Returns (2013) — link
  42. The Network of Firms Implied by the News (2020) — link
  43. The Current State of the Arbitrage Pricing Theory (1992)
  44. Information Markets and the Comovement of Asset Prices (2006)
  45. Understanding Contrastive Representation Learning Through Alignment and Uniformity on the Hypersphere (2020) — link
  46. Equilibrium in Economies with Incomplete Financial Markets (1985)
  47. C-Pack: Packed Resources for General Chinese Embeddings (2024) — link

3 of 55 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.