Ruin by dynamic contagion claims
Highlights
► Modelling of claims by a dynamic contagion process. ► Process generalises the Cox process and the Hawkes process. ► Asymptotics of ruin probabilities are discussed. ► A very efficient simulation procedure is implemented.
Introduction
In the classical Cramér–Lundberg risk model, the arrival of claims is modelled by a Poisson process. As substantially discussed in the literature, this model is often not realistic in practice and hence a variety of extensions have been studied. Many researchers, such as Björk and Grandell (1988) and Embrechts et al. (1993) had already suggested using the Cox process to model the arrival of claims, (see also the book by Grandel (1991)). Schmidli (1996) investigated the case for a Cox process with a piecewise constant intensity. More recently, Albrecher and Asmussen (2006) discussed a Cox process with shot noise intensity. On the other hand, only a few researchers have proposed risk models using self-excited processes, due to the observation of the clustering arrival of claims in reality, a similar pattern in the credit risk from the financial market, particularly during the current economic crisis. Stabile and Torrisi (2010) looked at the ruin problem in a model using the Hawkes process, a self-excited point process introduced by Hawkes (1971).
To capture the clustering phenomenon as well as some common external factors involved for the arrival of claims within one single consistent framework, in this paper, we extend further to use the dynamic contagion process introduced by Dassios and Zhao (2011), a generalisation of the externally excited Cox process with shot noise intensity (with exponential decay) and the self-excited Hawkes process (with exponential decay). It could be particularly useful for modelling the dependence structure of the underlying arriving events with dynamic contagion impact from both endogenous and exogenous factors. In this paper, we try to generalise results obtained for the classical model.
We organise our paper as follows. Section 2 provides distributional results we will use, mainly developed in Dassios and Zhao (2011). Section 3 formulates the problem. It also provides a numerical example and some asymptotics that are based on simulations. In Section 4, we use the martingale method and generalise Lundberg’s fundamental equation. We derive bounds for the ruin probability in Section 5. In Section 6, we derive all results via a change of measure. This makes simulations more efficient as ruin is certain under the new measure. Section 7 concentrates on exponentially distributed claims. Our results are illustrated by a numerical example.
Section snippets
Dynamic contagion process
The dynamic contagion process includes both the self-excited jumps (which are distributed according to the branching structure of a Hawkes process with exponential fertility rate) and the externally excited jumps (which are distributed according to a particular shot noise Cox process). We directly use the definition of the dynamic contagion process from Dassios and Zhao (2011). Definition 2.1 Dynamic Contagion Process The dynamic contagion process is a cluster point process on : The number of points in the time interval is
Ruin problem
We consider a company with its surplus process in continuous time on a probability space , where
- •
is the initial reserve at time ;
- •
is the constant rate of premium payment per time unit;
- •
is a point process () counting the number of cumulative arrived claims in the time interval , driven by a dynamic contagion process with its stochastic intensity process and the initial intensity ;
- •
is a sequence of independent identical
Exponential martingales and generalised Lundberg’s fundamental equation
In this section, we find some useful exponential martingales which link to the generalised Lundberg’s fundamental equation. More importantly, they are crucial for deriving some key results of the ruin problem in the later sections. Theorem 4.1 Assume and the net profit condition (6), we have a martingalewhere constants and satisfy a generalised Lundberg’s fundamental equationIf , then(7) has a
Ruin probability via original measure
Theorem 5.1 The ruin probability conditional on and is given by
Proof By the optional stopping theorem, a bounded martingale stopped at a stopping time is still a martingale. Now we consider the martingale found by Theorem 4.1 stopped at the ruin time, i.e. By the martingale property, we have and
Ruin probability via change of measure
In this section, we investigate the ruin probability and asymptotics by change of measure via the martingale derived by Theorem 4.1. We will find that under this new measure the ruin becomes certain, and this makes the simulation more efficient than under the original measure where the ruin is not certain and even rare. Similar ideas of improving simulation of rare events by change of measure can also be found in Asmussen (1985) and more recently Asmussen and Glynn (2007).
Example: jumps with exponential distributions
To represent the previous results in explicit forms, in this section, we further assume the externally excited and self-excited jumps in the intensity process and the claim sizes all follow exponential distributions, i.e. and , with the density functions and the Laplace transforms
Acknowledgment
We would like to thank an anonymous referee for various helpful suggestions.
References (13)
Conjugate processes and the simulation of ruin problems
Stochastic Processes and their Applications
(1985)- et al.
Ruin probabilities and aggregate claims distributions for shot noise Cox processes
Scandinavian Actuarial Journal
(2006) - et al.
Stochastic Simulation: Algorithms and Analysis
(2007) - et al.
Exponential inequalities for ruin probabilities in the Cox case
Scandinavian Actuarial Journal
(1988) - et al.
Martingales and Insurance Risk
Communications in Statistics–Stochastic Models
(1989) - et al.
A dynamic contagion process
Advances in Applied Probability
(2011)
Cited by (32)
An expansion formula for Hawkes processes and application to cyber-insurance derivatives
2023, Stochastic Processes and their ApplicationsExploring the contagion effect of social media on mass shootings
2022, Computers and Industrial EngineeringCitation Excerpt :Mohler et al. (2011) illustrated the implementation of the self-exciting point process model in urban crimes. Dassios & Zhao (2012) considered the risk process of claim arrival modeled by the dynamic contagion process. Rizoiu et al. (2018) established a novel connection between the epidemic model and the Hawkes point processes for online information modeling in geophysics and finance.
Hawkes processes in insurance: Risk model, application to empirical data and optimal investment
2021, Insurance: Mathematics and EconomicsOptimal reinsurance-investment strategy for a dynamic contagion claim model
2020, Insurance: Mathematics and EconomicsCitation Excerpt :This process includes both self-excited and externally-excited jumps and is extremely versatile for modelling purposes, allowing for a wide variety of features in the claim arrival dynamics (such as the frequency, magnitude of the impact, and the decay with time) to be captured. We refer the reader to Dassios and Zhao (2011) for more on this, as well as an analysis of a ruin problem in infinite time horizon in Dassios and Zhao (2012). In this paper, we propose to study the optimal reinsurance-investment problem in the framework of the claim contagion model introduced by Dassios and Zhao (2011).
Limit theorems for non-Markovian marked dynamic contagion processes
2018, Journal of Mathematical Analysis and ApplicationsModeling microstructure price dynamics with symmetric Hawkes and diffusion model using ultra-high-frequency stock data
2017, Journal of Economic Dynamics and Control