Paper: arXiv 2404.16467

Authors: Cecilia Aubrun, Rudy Morel, Michael Benzaquen, Jean-Philippe Bouchaud

Abstract

Cascades of events and extreme occurrences have garnered significant attention across diverse domains such as financial markets, seismology, and social physics. Such events can stem either from the internal dynamics inherent to the system (endogenous), or from external shocks (exogenous). The possibility of separating these two classes of events has critical implications for professionals in those fields. We introduce an unsupervised framework leveraging a representation of jump time-series based on wavelet coefficients and apply it to stock price jumps. In line with previous work, we recover the fact that the time-asymmetry of volatility is a major feature. Mean-reversion and trend are found to be two additional key features, allowing us to identify new classes of jumps. Furthermore, thanks to our wavelet-based representation, we investigate the reflexive properties of co-jumps, which occur when multiple stocks experience price jumps within the same minute. We argue that a significant fraction of co-jumps results from an endogenous contagion mechanism.

Complexity vs Empirical Score

  • Math Complexity: 6.5/10
  • Empirical Rigor: 6.0/10 (capped from 7.0: no out-of-sample test; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.0)

Why this score: The paper employs advanced wavelet-based representation and PCA for classification, involving moderate mathematical density, while the empirical work is highly data-driven with a large, specific dataset (301 US stocks over 8 years) and detailed methodology for jump detection and co-jump analysis.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Distinguish Endogenous vs Exogenous Price Jumps"] --> B["Methodology<br>Unsupervised Wavelet-Based Representation"]
  B --> C["Input Data<br>High-Frequency Stock Price Data"]
  C --> D["Compute Wavelet Coefficients<br>(Capture Multi-scale Dynamics)"]
  D --> E["Feature Extraction<br>Time-Asymmetry, Mean-Reversion, Trend"]
  E --> F["Clustering Analysis<br>Identify Distinct Jump Classes"]
  F --> G["Key Findings<br>1. New Jump Classes Discovered<br>2. Endogenous Contagion in Co-jumps"]
  style G fill:#d4edda,stroke:#155724

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 an unsupervised framework that uses wavelet scattering coefficients to embed jump time‑series for clustering and classification of price jumps and co‑jumps.

Source sentence
We introduce an unsupervised framework leveraging a representation of jump time-series based on wavelet coefficients and apply it to stock price jumps.

Assumes 5

  • Jumps occurring within the same time‑cluster after an initial jump are replicas of the initial jump. model
    Source sentence
    We assume that jumps taking place within the same “time-cluster” subsequent to an initial jump are merely replicas of the initial jump.
  • Jump detection uses a 4‑sigma deviation from zero as the threshold. statistical
    Source sentence
    The resulting method comes down to detecting price movements where the z-score deviates by more than 4-sigma from their average value (here equal to zero).
  • Under the null hypothesis of Gaussian residuals, the absolute jump‑score follows a Gumbel distribution. statistical
    Source sentence
    Under the null hypothesis of Gaussian residuals (no jump hypothesis) |x(t)| converges towards a Gumbel distribution.
  • A symmetric 2‑hour window (1 hour before and after the jump) captures the relevant price dynamics. data
    Source sentence
    we adopt a time window of 2 hours centered around the jump occurrence at time t = 0. Consequently, for each jump we extract a time-series of 119 rescaled returns x(t), corresponding to 1 hour preceding the jump and 1 hour following the jump.
Show 1 more
  • Only simultaneous jumps within the same minute are considered co‑jumps; lagged jumps are ignored. data
    Source sentence
    A co-jump is simply defined as a set of jumps occurring in the same minute. Here we avoid tackling the question of lagged jumps and consider only simultaneous jumps (up to the minute resolution).

Data

  • US stock price jumps dataset market-prices unknown
    January 2015 to December 2022 · 301 US stocks
    Source sentence
    We implement such detection on 301 US stocks from January 2015 to December 2022, considering only what happened between 10:30 and 15:00 in order to avoid special jumps due to the hig…
  • Synthetic time‑series for validation synthetic unknown
    Source sentence
    In order to validate the above analysis, we created synthetic time-series with volatility profiles of varying time-asymmetry and applied our classification method.

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • Approximately 50 % of jumps show positive volatility asymmetry, which would naïvely be labeled exogenous, but many are actually endogenous. statistical
    Source sentence
    Using the above classification, we find that a large proportion (∼50%) of our sample exhibit positive asymmetry and should naively be considered as exogenous jumps.
  • Co‑jump sizes follow a power‑law distribution with exponent close to 1, consistent with a near‑critical branching (contagion) process. statistical
    Source sentence
    the tail of this distribution is well fitted by a power-law S−τ with exponent τ ≈1, with a cut-off for S ≳100.
  • Three main classification directions are identified: volatility time‑asymmetry (endogenous, exogenous, anticipatory), mean‑reversion, and trend alignment. theoretical
    Source sentence
    Thanks to an unsupervised approach based on wavelet scattering coefficients, we have identified three main directions along which price jumps can be classified.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 15 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 2 papers in the archive

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 (39)

  1. The endogenous dynamics of markets: Price impact, feedback loops and instabilities (2011)
  2. Endogenous liquidity crises (2020)
  3. Aggregate fluctuations from independent sectoral shocks: self-organized criticality in a model of production and inventory dynamics (1993)
  4. Collective synchronization and high frequency systemic instabilities in financial markets (2018)
  5. High-frequency trading synchronizes prices in financial markets (2012) — link
  6. Stock price jumps: news and volume play a minor role (2008)
  7. May’s instability in large economies (2019)
  8. Temporal criticality in socio-technical systems (2023) — link
and 31 more
  1. Modelling systemic price cojumps with Hawkes factor models (2015)
  2. Exogenous and endogenous price jumps belong to different dynamical classes (2022)
  3. Invariant scattering convolution networks (2013)
  4. Scale Dependencies and Self-Similarity Through Wavelet Scattering Covariance (2022) — link
  5. Path Shadowing Monte-Carlo — → 2308.01486
  6. Endogenous versus exogenous origins of crises (2006)
  7. Endogenous versus exogenous shocks in systems with memory (2003)
  8. Robust dynamic classes revealed by measuring the response function of a social system (2008)
  9. Endogenous versus exogenous shocks in complex networks: An empirical test using book sale rankings (2004)
  10. Dynamics of book sales: Endogenous versus exogenous shocks in complex networks (2005)
  11. Critical reflexivity in financial markets: a Hawkes process analysis (2013)
  12. Branching-ratio approximation for the self-exciting Hawkes process (2014)
  13. Quantifying reflexivity in financial markets: Toward a prediction of flash crashes (2012)
  14. How Nature Works: The Science of Self-Organized Criticality (Copernicus, New York) (1996)
  15. Complexity, contingency, and criticality. (1995)
  16. Self-exciting point process modeling of crime (2011)
  17. Epidemiological modelling of the 2005 French riots: a spreading wave and the role of contagion (2018)
  18. Epileptic seizures: quakes of the brain? (2010)
  19. Prediction (2010)
  20. Robust estimation of intraweek periodicity in volatility and jump detection (2011)
  21. V-shapes (2022)
  22. Classification of endogenous and exogenous bursts in collective emotions based on Weibo comments during COVID-19 (2022)
  23. Detection and classification of high frequency transients using wavelet analysis (2002)
  24. A unified theory for steerable and quadrature filters (2007)
  25. Wavelet-based event detection method using PMU data (2015)
  26. Transient event classification based on wavelet neuronal network and matched filters (2018)
  27. Wavelet-based classification of transient signals for gravitational wave detectors (2018)
  28. Kernel principal component analysis (1997)
  29. Multivariate quadratic Hawkes processes—part I: theoretical analysis (2023)
  30. The Alchemy of Finance: Reading the Mind of the Market by George Soros (1994-05-06)
  31. The theory of branching processes (1963)

1 of 39 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.