Jointly Exchangeable Collective Risk Models: Interaction, Structure, and Limit Theorems
Daniel Gaigall, Stefan Weber · 2025
Read Jointly Exchangeable Collective Risk Models: Interaction, Structure, and Limit Theorems by Daniel Gaigall, Stefan Weber (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-06287.txt.
Introduction
Daniel Gaigall and Stefan Weber build a risk model for insurance portfolios where claims interact, in a 2025 paper. Classic models treat each policy's risk as separate, which misses the way losses can move together.
Their framework uses jointly exchangeable arrays, a symmetry notion broader than the classic one and described by a known representation theorem. Within it, they prove central limit theorems for the total losses of a portfolio. Those theorems justify simple approximations for large portfolios and long horizons. Computer experiments confirm the approximations behave as promised. The authors also study how dependence between risks shapes the loss distribution, with care for the heavy tail where ruin lives.
Catalog record
| CREATOR | Daniel Gaigall, Stefan Weber |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 74 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-06287
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Jointly Exchangeable Collective Risk Models: Interaction, Structure, and Limit Theorems Daniel Gaigall, Stefan Weber Abstract We introduce a framework for systemic risk modeling in insurance portfolios using jointly exchangeable arrays, extending classical collective risk models to account for interactions. Joint exchangeability is a more general probabilistic symmetric than de Finetti’s exchangeability, characterized by the Aldous-Hoover-Kallenberg representation. We establish central limit theorems that asymptotically capture total portfolio losses, providing a theoretical foundation for approximations in large portfolios and over long time horizons. These approximations are validated through simulation-based numerical experiments. Additionally, we analyze the impact of dependence on portfolio loss distributions, with a particular focus on tail behavior. † † E-Mail: [email protected] (Daniel Gaigall); [email protected] (Stefan Weber) Keywords: Central limit theorem; Collective risk model; Joint Exchangeability; Systemic risk; Contagion; Dependence; Insurance mathematics; Loss distributions 1 Introduction Systemic risk is a major concern in finance and insurance. Idiosyncratic and systematic risks account for only part of the probabilistic variations in outcomes, while systemic interactions play a crucial role in financial markets, cyber networks, power grids, and other infrastructures. This paper focuses on modeling insurance portfolio losses within a generalized collective model, or frequency-severity model. Classical risk theory does not inherently capture systemic effects, highlighting the well-recognized need for more sophisticated models. We introduce a probabilistic model for total losses, such as those of an insurance company, using the theory of jointly exchangeable arrays, a weaker notion than classical exchangeability. This approach extends beyond idiosyncratic and systematic risk by incorporating a specific systemic component: network interaction. Specifically, our framework generalizes de Finetti’s exchangeability by preserving network interactions while allowing for relabeling, unlike conventional models that assume homogeneity after adjusting for covariates. Example specifications include random graphs and graphon models – such as Erdős-Rényi random networks – to describe transmissions and infections within an insurance portfolio. Additionally, conditional loss expenditures are modeled as dependent through jointly exchangeable arrays. The classical collective risk model emerges as a special case. Within this framework, we establish several limit theorems that asymptotically characterize the distribution of total portfolio losses. In insurance, systemic risk naturally arises in networked systems. Relevant examples include power grids and digital infrastructures, operating both at local and global scales, as well as contagion-type dynamics generated by cyber threats. Market sentiment can likewise be modeled within this framework. Recent surveys addressing these phenomena include Awiszus et al. (2023) and ENISA (2024) . Our approach is most closely related to the systemic perspective of Zeller & Scherer (2022) , while placing particular emphasis on underlying probabilistic symmetries and on the derivation of corresponding limit theorems. For a systematic presentation of probabilistic symmetries, including joint exchangeability, we refer to Aldous et al. (1985) and Kallenberg (2005) . Under the additional assumption that jointly exchangeable arrays are dissociated, limit results in distribution for related statistics are derived in Silverman (1976) , with further results in Eagleson & Weber (1978) . Jointly exchangeable distributions admit an ergodic decomposition that extends de Finetti’s theorem, see de Finetti (1937) . These structural results are developed in Hoover (1979) , Aldous (1981) , Kallenberg (1988) , Kallenberg (1989) , and Kallenberg (2005) . The Choquet representation of the joint distribution as a mixture of dissociated arrays provides a natural reduction to this case. Limit theorems for exchangeable arrays have been studied from various perspectives. Barbour & Eagleson (1985) examines dissociated arrays and generalizations of central limit theorems to U -statistics. Davezies, D’Haultfœuille & Guyonvarch (2021) extends these results to empirical processes. Austern & Orbanz (2022) develops a general theory of limit theorems within the framework of distributional symmetries defined by invariance under group transformations. Although this broader theory applies in our setting, we use the classical results of Silverman (1976) and Eagleson & Weber (1978) , which are sufficient for our analysis. Building on these theoretical foundations, we now turn to the implications for risk management and loss distribution modeling. Of particular interest are the properties of the underlying loss distribution. In our generalized collective model with interacting losses, the expectation and variance of the total loss of the insurance company can be explicitly computed. We establish asymptotic characterizations of the total loss distribution as the number of contracts or the time horizon grows. The central limit theorems we establish provide a useful tool for risk management and include Anscombe-type results; see Anscombe (1952) . Simulations demonstrate that these asymptotic approximations remain accurate even for moderate portfolio sizes and relatively short time horizons. The use of probabilistic interacting particle models to capture systemic risk in economic applications dates back to Föllmer (1974) . In credit risk, Giesecke & Weber (2004) ; Giesecke & Weber (2006) asymptotically characterize total portfolio losses under local interaction, specifically employing a voter model. In insurance, the collective model with systemic effects is particularly relevant for cyber risk and associated losses. For surveys on cyber insurance, see Awiszus et al. (2023) , Dacorogna & Kratz (2023) , Eling (2020) , ENISA (2024) , and Knispel et al. (2024) . Closely related to our approach, Zeller & Scherer (2022) study a cyber collective model incorporating covariates but without network interactions. Other forms of local and global interactions in actuarial cyber risk models are studied in Fahrenwaldt, Weber & Weske (2018) , Hillairet & Lopez (2021) , Hillairet et al. (2022) , Bessy-Roland, Boumezoued & Hillairet (2021) , Hillairet, Réveillac & Rosenbaum (2023) , and Awiszus et al. (2024) , though without addressing limit theorems under joint exchangeability. The main contributions of this paper are (i) a novel framework for systemic risk modeling in insurance portfolios based on jointly exchangeable arrays, extending classical collective risk models to incorporate interactions, (ii) asymptotic characterizations of total losses via central limit theorems, providing theoretical justification for approximations in large portfolios and over long time horizons, (iii) simulation-based validation of the model and its approximations, demonstrating the practical relevance of the approach for risk management applications. The structure of the paper is as follows. Section 2 introduces the extended collective model, discusses joint exchangeability, and provides illustrative examples. Section 3 presents limit theorems for total losses and characterizes their relation to classical collective models. All proofs are deferred to Appendix A . In Section 4 , we conduct case studies, validate the accuracy of approximations, and analyze the impact of dependence. Finally, Section 5 concludes with a discussion of open research questions. 2 A collective loss model with interaction In this paper, we are interested in a frequency-severity model of insurance losses, also called collective model, which includes a special form of contagious interactions. We will first explain the type of interaction on which we focus before, we describe the full dynamic model. The underlying probability space will be denoted by (\Omega,\mathfrak{A},P) and assumed to be atomless. 2.1 Contagious losses We consider a collection of agents or, in a collective model, of loss occurrences, enumerated by the natural numbers \mathbb{N} . In the latter case, unlike some common conventions, losses of zero are also admissible. Throughout the paper, the terms agent and entity will be used interchangeably. We assume that agents can trigger losses of others, a phenomenon that can be captured by a graph as illustrated in Figure 1 . Vertices represent entities,