Paper: arXiv 2610.14948

Authors: Sayak Dutta

Abstract

Temporal point processes are used both to measure market mechanisms, for example how fast prices absorb news and how much of the reaction is the market responding to itself, and to predict the next event. We show that one algebraic object, the phase-type (Erlang-chain) state space, serves both purposes with exact likelihoods. For measurement, we build a multiscale phase-type Hawkes process whose parameters all enter linearly. Its log-likelihood is concave, and we compute the global maximum together with a Fenchel duality-gap certificate. A Markov embedding gives the response to an exogenous shock, echoes included, in closed form, and this response is stable if and only if the spectral radius of the branching matrix is below one. Applied to six markets around 670 scheduled macroeconomic releases (2022-2026) with a built-in placebo test, the model shows that the direct reaction to CPI, payrolls, PPI and the FOMC statement is absorbed in about a second, while the market’s own echoes stretch half of the response to 16-137 s. For prediction, we introduce EPT-X, a neural point process that adds a history-dependent mixture-of-Erlang renewal channel to a linear Hawkes backbone. Its intensity and compensator remain in closed form, its states follow an affine recursion that trains with a parallel scan, and it can start from the certified fit of the measurement model. EPT-X has the best test likelihood on all five LOBSTER order-book streams and on Retweet and Taobao, and is level with the best published model on StackOverflow. We also find that the EasyTPP implementation of IntensityFree clamps short inter-event times before scoring, and that the gaps of the EasyTPP Amazon data are uniform on two fixed intervals.

Complexity vs Empirical Score

  • Math Complexity: 9.0/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: This paper presents a highly mathematical and novel approach to modeling event streams using phase-type point processes, demonstrating strong empirical validation across multiple datasets. The methodology is well-defined, and the results are significant for both measurement and prediction in quantitative finance and machine learning.

Research Flowchart

  flowchart TD
    A[Research Goal: Certified Measurement & Prediction of Event Streams] --> B{Key Methodology: Exact Phase-Type Point Processes};
    B -- Measurement --> C[Multiscale Phase-Type Hawkes Process: Linear Parameters, Concave Likelihood, Fenchel Duality];
    B -- Prediction --> D[EPT-X: Neural Point Process, Erlang Renewal Channel, Closed-Form Intensity/Compensator];
    C -- Data/Inputs --> E[6 Markets, 670 Macroeconomic Releases (2022-2026), Placebo Test];
    D -- Data/Inputs --> F[5 LOBSTER Order-Book Streams, Retweet, Taobao, StackOverflow];
    E -- Computational Process --> G[Global Maximum Likelihood, Markov Embedding for Response];
    F -- Computational Process --> H[Affine Recursion, Parallel Scan Training, Certified Fit from Measurement Model];
    G --> I[Key Findings (Measurement): Direct Reaction ~1s, Market Echoes 16-137s];
    H --> J[Key Findings (Prediction): Best Test Likelihood on LOBSTER, Retweet, Taobao; Level with Best on StackOverflow; EasyTPP Issues Identified];

Cited References (55)

  1. Spectra of some self-exciting and mutually exciting point processes (1971) — link
  2. Stability of nonlinear H awkes processes (1996) — link
  3. Micro effects of macro announcements: Real-time price discovery in foreign exchange (2003) — link
  4. Modeling foreign exchange market activity around macroeconomic news: H awkes-process approach (2015) — link
  5. Critical reflexivity in financial markets: a H awkes process analysis (2013) — link
  6. Quantifying reflexivity in financial markets: Toward a prediction of flash crashes (2012) — link
  7. Apparent criticality and calibration issues in the H awkes self-excited point process model: application to high-frequency financial data (2015) — link
  8. H awkes process model with a time-dependent background rate and its application to high-frequency financial data (2017) — link
and 47 more
  1. H awkes processes in finance (2015) — link
  2. Estimation of slowly decreasing H awkes kernels: application to high-frequency order book dynamics (2016) — link
  3. Uncovering causality from multivariate H awkes integrated cumulants (2018)
  4. tick: a P ython library for statistical learning, with an emphasis on H awkes processes and time-dependent models (2018)
  5. Learning social infectivity in sparse low-rank networks using multi-dimensional H awkes processes (2013)
  6. Estimation of space–time branching process models in seismology using an EM -type algorithm (2008) — link
  7. An estimation procedure for the H awkes process (2017) — link
  8. Simple and globally convergent methods for accelerating the convergence of any EM algorithm (2008) — link
  9. Matrix Iterative Analysis (2000) — link
  10. The time-rescaling theorem and its application to neural spike train data analysis (2002) — link
  11. Stochastic Models: An Algorithmic Approach (1994)
  12. Modeling and evaluating insurance losses via mixtures of E rlang distributions (2010) — link
  13. Recurrent marked temporal point processes: Embedding event history to vector (2016) — link
  14. The neural H awkes process: A neurally self-modulating multivariate point process (2017)
  15. Fully neural network based model for general temporal point processes (2019)
  16. Intensity-free learning of temporal point processes (2020)
  17. Transformer H awkes process (2020)
  18. Self-attentive H awkes process (2020)
  19. Neural temporal point processes: A review (2021) — link
  20. Transformer embeddings of irregularly spaced events and their participants (2022)
  21. On the predictive accuracy of neural temporal point process models for continuous-time event data (2023)
  22. EasyTPP : Towards open benchmarking temporal point processes (2024)
  23. Mamba H awkes process (2024) — link
  24. Deep continuous-time state-space models for marked event sequences (2024) — link
  25. Hyper H awkes processes: Interpretable models of marked temporal point processes (2025) — link
  26. LOBSTER : Limit order book reconstruction system (2011) — link
  27. Decomposable Transformer Point Processes (2024)
  28. Attention is all you need (2017)
  29. Learning phrase representations using RNN encoder–decoder for statistical machine translation (2014) — link
  30. Decoupled weight decay regularization (2019)
  31. High frequency lead/lag relationships—Empirical facts (2014) — link
  32. Modeling financial contagion using mutually exciting jump processes (2015) — link
  33. A cluster process representation of a self-exciting process (1974) — link
  34. An Introduction to the Theory of Point Processes: Volume I : Elementary Theory and Methods (2003) — link
  35. On L ewis’ simulation method for point processes (1981) — link
  36. Non-parametric kernel estimation for symmetric H awkes processes. A pplication to high frequency financial data (2012) — link
  37. First- and second-order statistics characterization of H awkes processes and non-parametric estimation (2016) — link
  38. Hawkes graphs (2018) — link
  39. Limit theorems for nearly unstable H awkes processes (2015) — link
  40. Learning G ranger causality for H awkes processes (2016)
  41. Fitting phase-type distributions via the EM algorithm (1996)
  42. Stability, convergence to equilibrium and simulation of non-linear H awkes processes with memory kernels given by the sum of E rlang kernels (2019) — link
  43. Neural jump stochastic differential equations (2019)
  44. Neural spatio-temporal point processes (2021)
  45. Simplified state space layers for sequence modeling (2023)
  46. Advances in temporal point processes: B ayesian, neural, and LLM approaches (2026)
  47. HoTPP benchmark: Are we good at the long horizon events forecasting? (2024) — link

0 of 55 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.