Precise large deviations for a multidimensional risk model with regression dependence structure

Yang Liu, Ke-Ang Fu, Zhenlong ChenProbability 2023 · DOI

Abstract

This paper investigates a nonstandard multidimensional risk model where claim sizes are represented as independent and identically distributed (i.i.d.) random vectors with dependent components. A key innovation is the introduction of a regression dependence structure between inter-arrival times and claim-size vectors. Under the assumption that univariate marginal distributions of claim vectors possess consistently varying tails, the authors derive precise large deviation formulas. These results provide a robust framework for estimating extreme loss probabilities in complex, multi-line insurance portfolios.

Introduction

In classical risk theory, the independence between claim sizes and their arrival times is a standard simplifying assumption. However, empirical evidence suggests that these variables are often coupled. For instance, a high-frequency period of small claims might be followed by a significant delay before a major catastrophic event. Furthermore, modern insurers typically manage multiple business lines simultaneously, where a single underlying cause can trigger claims across different sectors (e.g., property and casualty). This paper extends the literature by modeling these dependencies through a multidimensional framework incorporating a regression dependence structure.

Problem Statement and Model Formulation

The study addresses the challenge of calculating "precise large deviations"—the probability of rare, catastrophic total losses—in a multidimensional setting. The model considers an aggregate claim amount vector $\vec{S}(t)$, which is the sum of claim vectors over time.

The core complexity lies in the regression dependence structure. Unlike models where an inter-arrival time $\theta_n$ depends only on the immediate next claim $X_n$, this model defines a $d$-dependent sequence. Here, the distribution of $\theta_n$ is conditioned on a window of previous claim vectors $(\vec{X}_{n-d}, \dots, \vec{X}_n)$. This captures a "memory" effect within the insurance portfolio, where past claim severities influence the timing of future events.

Methodology

The authors employ advanced probabilistic techniques to handle the mathematical difficulties of dependent multidimensional processes:

  1. Heavy-Tail Modeling: Claim sizes are assumed to follow distributions with "consistently varying tails" (the $\mathcal{C}$ class). These fat-tailed distributions are essential for modeling insurance risks where extreme values are more frequent than predicted by Gaussian models.
  2. Multidimensional Compound Sums: The aggregate risk is modeled as a vector process to account for dependencies between different insurance lines.
  3. Stochastic Bounding: To derive asymptotic results, the authors utilize a "generalized multi-delayed renewal counting process." This allows them to bound the complex dependent arrival process using simpler, manageable variables, facilitating the proof of the main theorem.

Key Results

The primary contribution of this research is the derivation of a precise asymptotic formula for the probability that the aggregate claim amount exceeds its expected value by a large margin $\vec{x}$ as time $t \to \infty$:

$$P(\vec{S}(t) - \vec{\mu}\lambda t > \vec{x}) \sim (\lambda t)^m \prod_{i=1}^m \bar{F}_i(x_i)$$

This result demonstrates that despite the intricate regression dependence between timing and size, the asymptotic behavior of extreme risks is primarily driven by the heavy-tailed nature of the claims themselves. The formula indicates that the probability of simultaneous extreme losses across all $m$ lines is proportional to the product of individual tail probabilities, scaled by the expected claim frequency $(\lambda t)$.

Discussion and Conclusion

The findings suggest that for heavy-tailed risks, the specific structure of regression dependence does not fundamentally alter the tail behavior of the aggregate loss, provided a strong law of large numbers holds for the arrival process. This provides actuaries with a simplified yet theoretically grounded method for estimating insolvency risk in complex environments.

Future research may explore non-stationary arrival processes to account for environmental or economic shifts and extend the framework to broader classes of heavy-tailed distributions, such as subexponential tails.

Summary updated

Abstract

Abstract In this paper, we consider a nonstandard multidimensional risk model, in which the claim sizes $\{\vec{X}_k, k\ge 1\}$ form an independent and identically distributed random vector sequence with dependent components. By assuming that there exists the regression dependence structure between inter-arrival time and the claim-size vectors, we extend the regression dependence to a more practical multidimensional risk model. For the univariate marginal distributions of claim vectors with consistently varying tails, we obtain the precise large deviation formulas for the multidimensional risk model with the regression size-dependent structure.
BibTeX
@inproceedings{liu2023precise,
  author = {Yang Liu and Ke-Ang Fu and Zhenlong Chen},
  title = {{Precise large deviations for a multidimensional risk model with regression dependence structure}},
  booktitle = {Probability},
  year = {2023},
  doi = {10.1017/s0269964823000220},
}
Figure