Paper: arXiv 2610.05741

Authors: Sebastián Calcetero Vanegas, Ian Weng Chan

Abstract

Credibility theory combines individual experience with portfolio information for insurance pricing, but classical formulations focus primarily on conditional means and expected premiums. We propose a Dirichlet mixed-membership model (DMMM) for multivariate distributional credibility. Policyholder risk is represented by a stable composition over latent risk classes: baseline characteristics determine its a priori assessment, while repeated multivariate experience progressively updates its a posteriori assessment. Support-specific expert distributions accommodate heterogeneous outcomes, with dependence induced through the shared latent structure. The resulting posterior predictive distribution can be written exactly as a convex combination of portfolio and experience components, extending the familiar credibility-factor structure beyond the conditional mean. We study the framework through a simulation experiment and a vehicle telematics application. The simulation shows that relatively simple experts can capture complex multivariate insurance distributions and learn policyholder-specific risk as experience accumulates. In the telematics application, recent claims and driving behaviour provide complementary information for future claim-frequency prediction, allowing similar policyholders to receive different experience-rated assessments. The DMMM therefore provides a flexible and interpretable framework for combining heterogeneous insurance experience while retaining the structure of classical credibility.

Complexity vs Empirical Score

  • Math Complexity: 8.5/10
  • Empirical Rigor: 7.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: This paper introduces a novel and mathematically sophisticated Dirichlet mixed-membership model for multivariate distributional credibility, extending classical credibility theory. It demonstrates strong empirical rigor through both simulation and a real-world telematics application, providing a flexible and interpretable framework for insurance pricing.

Research Flowchart

  flowchart TD
    A["Research Goal: Extend Credibility Theory for Exact Multivariate Distributional Credibility"] --> B{"Key Methodology: Dirichlet Mixed-Membership Model (DMMM)"};
    B --> C["Data/Inputs: Portfolio Information, Individual Policyholder Experience (Multivariate), Baseline Characteristics, Support-Specific Expert Distributions"];
    C --> D{"Computational Process: Bayesian Inference for DMMM, Posterior Predictive Distribution Calculation"};
    D --> E["Outcomes: Exact Convex Combination of Portfolio & Experience Components (Credibility Factor Structure)"];
    E --> F["Key Findings: Model Captures Complex Multivariate Distributions, Learns Policyholder-Specific Risk (Simulation), Complementary Information from Telematics for Claim Prediction, Flexible & Interpretable Framework"];

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 Dirichlet mixed‑membership model (DMMM) for multivariate distributional credibility.

Source sentence
We propose a Dirichlet mixed-membership model (DMMM) for multivariate distributional credibility.

Assumes 5

  • Policyholder risk is represented by a stable composition over latent risk classes that does not change over time. model
    Source sentence
    The composition remains fixed over the observation horizon, while different periods may express different classes and the observed history progressively reveals the underlying composition.
  • Baseline characteristics determine the a priori (portfolio‑level) classification of the Dirichlet composition. model
    Source sentence
    Baseline attributes determine the portfolio‑level assessment of this composition, while different periods may express different classes and repeated observations progressively refine the insurer’s posterior assessment of the individual composition.
  • The Dirichlet prior on the class‑composition vector has strictly positive parameters. statistical
    Source sentence
    the Dirichlet prior has strictly positive parameters.
  • Expert distributions (class‑specific predictive models) are support‑specific and can accommodate heterogeneous outcomes. model
    Source sentence
    Support‑specific expert distributions accommodate heterogeneous outcomes, with dependence induced through the shared latent structure.
Show 1 more
  • The expert dictionary can be made arbitrarily rich (KL‑rich) so that the KL divergence to the true distribution vanishes as the number of experts grows. statistical
    Source sentence
    → ∞). This is stronger than weak denseness, but can be ensured by suitable support‑consistent expert families under standard regularity conditions.

Data

  • Synthetic portfolio of So et al. (2021) synthetic unknown
    Source sentence
    claims and telematics using the synthetic portfolio of So et al. (2021), extended to a controlled longitudinal setting motivated by the SEAT framework
  • Untouched test portfolio (telematics application) synthetic unknown
    Source sentence
    Using an untouched test portfolio, the proposed model attained the lowest test MSE in both persistence settings and the best deviance and predictive log score among the probabilistic models

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • The DMMM achieved the lowest test mean‑squared error (MSE) among all models in both moderate and high persistence settings. predictive-accuracy
    Source sentence
    When each credibility model is supplied with the complete two‑year history, the selected DMMM has the lowest test MSE point estimate in both persistence settings
  • The DMMM obtained the best mean Poisson deviance and the highest predictive log‑score among the probabilistic models. predictive-accuracy
    Source sentence
    It also has the lowest mean Poisson deviance and the best predictive log score among the probabilistic models
  • Linear credibility remained competitive for point prediction but did not provide a full predictive distribution and produced negative predictions in the moderate‑persistence cohort. negative
    Source sentence
    for point prediction and has the lowest MAE in both settings, but it does not define a full predictive count distribution and produces three negative predictions in the moderate‑persistence cohort
  • Calibration ratios were close to one for the common Poisson, DMMM, and linear models under high persistence, while the Gaussian random‑effects model over‑predicted the portfolio mean. statistical
    Source sentence
    Calibration is close to one for the common Poisson, DMMM, and linear models under high persistence, while the Gaussian model overpredicts the portfolio mean
  • Simulation results show that richer collections of risk classes (K≈12–20) sharply improve approximation quality, with the K=36 model providing further but smaller gains while remaining parsimonious. theoretical
    Source sentence
    Approximation quality improves sharply as the expert dictionary becomes richer, with most of the improvement occurring by K = 12–20. The K = 36 model provides a further, although smaller, reduction in joint KL divergence while remaining relatively parsimonious

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 45 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

https://github.com/ianwengchan/DMMM.jl — the authors' own repository

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. Experience Rating and Credibility (1967) — link
  2. A Course in Credibility Theory and Its Applications (2005)
  3. Glaubw "u rdigkeit f "u r Schadens "a tze (1970)
  4. Severity Estimation in Dependent Collective Risk Models (2026) — link
  5. Credibility Estimation of Distribution Functions with Applications to Experience Rating in General Insurance (2015) — link
  6. Effective Experience Rating for Large Insurance Portfolios via Surrogate Modeling (2024) — link
  7. Stan: A Probabilistic Programming Language (2017) — link
  8. Bayesian Credibility under a Bivariate Prior on the Frequency and the Severity of Claims (2021) — link
and 31 more
  1. Data Mining of Telematics Data: Unveiling the Hidden Patterns in Driving Behavior (2025) — link
  2. Joint Modeling of Claim Frequencies and Behavioral Signals in Motor Insurance (2022) — link
  3. Multivariate Credibility Modelling for Usage-Based Motor Insurance Pricing with Behavioural Data (2019) — link
  4. Multivariate Latent Risk: A Credibility Approach (2008) — link
  5. Multivariate Credibility for Aggregate Loss Models (2003) — link
  6. A Class of Mixture of Experts Models for General Insurance: Theoretical Developments (2019) — link
  7. The Added Value of Dynamically Updating Motor Insurance Prices with Telematics Collected Driving Behavior Data (2022) — link
  8. Credible Means Are Exact Bayesian for Exponential Families (1974)
  9. The Use of Collateral Data in Credibility Theory: A Hierarchical Model (1975)
  10. Modelling Compositional Data Using Dirichlet Regression Models (2009)
  11. DirichletReg : Dirichlet Regression for Compositional Data in R (2014)
  12. Bayesian Regression for a Dirichlet Distributed Response Using Stan (2018) — link
  13. PyMC : A Modern and Comprehensive Probabilistic Programming Framework in Python (2023) — link
  14. The Credibility Approach to Experience Rating (1979)
  15. Synthetic Dataset Generation of Driver Telematics (2021) — link
  16. Simulation Engine for Adaptive Telematics Data (2025) — link
  17. Weekly Dynamic Motor Insurance Ratemaking with a Telematics Signals Bonus–Malus Score (2025) — link
  18. Limiting Behavior of Posterior Distributions When the Model Is Incorrect (1966) — link
  19. Hierarchical Insurance Claims Modeling (2008) — link
  20. Misspecification in Infinite-Dimensional Bayesian Statistics (2006) — link
  21. A Flexible Bayesian Nonparametric Model for Predicting Future Insurance Claims (2017) — link
  22. Ascent-Based Monte Carlo Expectation–Maximization (2005) — link
  23. Optimal Bonus–Malus Systems Using Finite Mixture Models (2014) — link
  24. On Bayesian Mixture Credibility (2006) — link
  25. Nonparametric Bayesian Credibility (2009)
  26. Bayesian Ratemaking under Dirichlet Process Mixtures (2017) — link
  27. On Bayesian Credibility Mean for Finite Mixture Distributions (2024) — link
  28. Mixture Credibility Formulas (2026) — link
  29. Frequency-Severity Experience Rating Based on Latent Markovian Risk Profiles (2022) — link
  30. Improving Risk Classification and Ratemaking Using Mixture-of-Experts Models with Random Effects (2023) — link
  31. Applications of Population Sampling to Insurance Ratemaking and Reserving (2024) — link

0 of 39 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2601 linked papers.