On a Compound Poisson Risk Model Perturbed by Brownian Motion with Variable Premium and Tail Dependence between Claims Amounts and Inter-Claim Time ()
1. Introduction
For a long time, risk modeling in actuarial science was based on the compound Poisson risk model introduced by the Swedish actuary Filip Lundberg (1903) and relayed by Harald Cramér (1930). This model in its original version is based on the independence between the random variables involved in the risk model (see for example [1] - [6] ). However, in certain practical contexts, this assumption is inadequate and too restrictive. For instance, in flood insurance, the occurrence of multiple floods in a short period can lead to significant damages and claim amounts due to the accumulation of water. In earthquake insurance, it is the opposite, as in a high-risk area, the longer the time between two earthquakes, the more significant the second earthquake due to the accumulation of energy. This observation will prompt actuarial investigations in order to take this dependence into account. Numerous research studies have considered a dependence between claims amounts and inter-claim time via Farlie-Gumbel-Morgenstern copula and have produced interesting results (see for example [7] - [16] ). This copula although commonly used in the literature, has certain limitations. It fails to model tail dependencies because it has an upper and lower tail dependence coefficient of zero. Based on this observation, authors will explore other copulas to express tail dependencies. Many works based on Spearman copula have produced satisfactory results on this subject (see for example [17] - [22] ). In this work, this copula is retained as a tool for the dependence structure.
The actuarial industry is increasingly confronted with multiple, complex and varied situations. This has led many authors to incorporate a perturbation component via Brownian motion into the risk model in order to better reflect the growth pattern of insurance (see for example [23] - [31] ).
In order to enable insurance companies to face the challenges raised above and remain competitive in the insurance market, scholars have found that it’s important to vary the premium according to the intensity of claims in the portfolio. Thus, Zhong Li and Kristina P. Sendova studied the classical risk model with variable premium and dependence between claims amounts and inter-claim times via Farlie-Gumbel-Morgenstern copula and without a disturbance component (see [32] ). Ying Shen studied the same model by adding a disturbance component through Brownian motion (see [33] ). This work is intended to be a continuation of the approach adopted by Ying Shen but taking into account the tail dependence via the Spearman copula. In this approach, two states govern the policyholders in the portfolio: the first class where the amount of claims is high, and the second where the claim amounts accumulate slowly, which motivates the use of the Spearman copula to capture tail dependencies. The premiums paid by policyholders in a given class are homogeneous but different from one class to another class and depend on random thresholds
. It is assumed that in the first class, the premium paid per unit of time is a constant denoted
and in the second, the premium paid per unit of time is a constant denoted
and
.
Denoting by
the surplus process (with
) and by
the premium collected by the insurer until time t, in the Compound Poisson perturbed by Brownian motion with variable premium and dependence between claims amounts and inter-claim times via Spearman copula, the model follows the following dynamic:
(1.1)
where:
is the surplus process (with
the initial surplus and
);
represents the premium collected by the insurer until time t and is a piecewise premium with
, the constant premium rate collected by the insurer per unit of time for each class for claims;
is the aggregated loss process with a compound Poisson distribution where:
*
is the total number of recorded claims up to time t, following a Poisson process (Note that
if
);
*
is a sequence of random variables representing the individual claim amount with a common density function f and cumulative distribution function F assumed to follow an exponential distribution with parameter β and mean μ;
*It is assumed that the distribution of the waiting time until the next claim depends on the amount of the previous claim via random threshold
,
. It is also assumed that the sequence
is a set of independent identical distribution random thresholds with common cumulative distribution
, probability density
and independent of the amount
of claims.
represents two classes of policyholders so that if the amount of claims
is such that
, the policyholders are placed in the first class and the waiting time
until the next claim follows an exponential distribution with parameter
and probability density
. If
, the policyholders are placed in the second class and the waiting time
until the next claim follows an exponential distribution with parameter
and probability density
;
.
is a standard Brownian motion independent with the aggregate claims process
and
is the diffusion volatility;
is an indicator function, which equals 1 if event A occurs and 0 otherwise.
The structure of dependence between premium and the number of claims mentioned in the risk model defined by the relation (1.1) was introduced by Albrecher and Boxman (2004) where the ultimate-survival probabilities are considered (voir [32] ).
The purpose of this work is to determine the integro-differential equations satisfied by Gerber-Shiu function and their Laplace transform in the risk model defined by Equation (1.1). To achieve this, the rest of the article is organized as follows: In Section 2, the preliminaries related to the risk model defined by Equation (1.1) will be presented. In Section 3, the integro-differential equations satisfied by the Gerber-Shiu function in the risk model defined by Equation (1.1), are determined. Section 4 is devoted to the study of the Laplace transform for Gerber-Shiu function of the risk model defined by the relation (1.1).
2. Preliminaries
2.1. Instant of Ruin and Ruin Probability
Let T be the instant of ruin of the insurance company. T is defined by:
(2.1)
When the probability of ruin is always zero, by convention, we denote
and in this case,
The probability of ruin other a finite time horizon t is defined by:
(2.2)
Similarly, the ultimate ruin probability is defined by:
(2.3)
2.2. Expected Discounted Penalty Function of Gerber-Shiu
The expected discounted penalty function of Gerber-Shiu, first appeared in the work of Gerber and Shiu in 1998 (see [5] ). Nowadays, this function is of significant interest in research.
Its analysis remains a central question both in insurance and finance, as it is a valuable tool not only for studying the probability of ruin but also calculating retirement and reinsurance premiums, option pricing, and more. In the risk model defined by the relation (1.1), this function is defined by:
(2.4)
where
is the instant of ruin defined by Equation (2.1);
is the instant just before ruin;
δ is a interest force;
The penalty function
is a positive function of the surplus just before ruin,
and the deficit at ruin
;
is a indicator function, wich equal 1 if event A occurs and 0 otherwise.
Because of the perturbation term, the expected discounted penalty function of the Gerber-Shiu is decomposed according to wether ruin is caused by claims or oscillation, i.e.
(2.5)
Where:
is the Gerber-Shiu when ruin is caused by claims and is defined by:
(2.6)
is the Gerber-Shiu when ruin is due to oscillation and is defined by:
(2.7)
with
, where:
and
are defined in the relation (2.4).
To guarantee that ruin will not be a certain event, the model must verify the following solvency condition:
(2.8)
2.3. Dependence Structure
2.3.1. Copulas
Copulas introduced by Abe Sklar in 1998, are an innovative and relevant tool for introducing dependence between multiples random variables. Given the marginal distribution functions of several random variables, copulas allow us to establish their joint distribution function. Nowadays, they are fundamental in modeling multivariate distributions in finance, insurance and hydrology. Key references on copulas theory include Joe [7] and Nelsen [13] .
Definition 2.1. A bivariate copula C is a non-decreasing, right-continuous function defined from
into
and satisfying the following properties:
1)
and
;
2)
and
;
3)
such that
and
, C verifies
.
Theorem 2.1. (Sklar’s theorem). Let two random variables
and
and F their joint distribution function with
and
their marginal. Then, there exists a copula C defined from
into
such that for all
and
in
,
.
2.3.2. Tail Dependence Concept
The concept of tail dependence relates to the amount of dependence in the upper-right-quadrant tail or lower-left-quadrant tail of a bivariate distribution. It is a concept that is relevant for the study of dependence between extreme value.
Definition 2.2. If is a bivariate copula such that the limit:
exists, then C has an upper tail dependence if
and has upper tail independence if
.
exists, then C has a lower tail dependence if
and has lower tail independence if
.
The real numbers
and
are called tail dependence coefficients.
Remark 2.1.
From a probabilistic point of view,
.
The Farlie-Gumbel-Morgenstern copula defined for all
by
, where
is a dependency parameter, has the tail dependence coefficients
and
, hence its inability to measure tail dependencies.
2.3.3. Model of Dependence Based on Spearman Copula
In this article, the structure of dependence is ensured by the Spearman copula. Il is defined for all
and
as follow:
(2.9)
where:
;
and α is a dependency parameter.
The Spearman copula allows for the introduction of positive dependence as well as tail dependencies in many situations. It is suitable for modelling dependence on extreme values because
and
. It also includes independence when
. Using Formula (2.9), the random vectors of claim amounts and inter-claim times
has the joint distribution function given by
(2.10)
where
are the marginal distributions of the random variables X and
, respectively.
The support of the copula
is
(see Nelsen [13] ).
On
,
and on D,
is the uniform distribution.
Since the dependence structure between the random variables X and
(
) is described by the copula
, they are monotonic and there is almost certainly an increasing function l such that
(see Nelsen [13] , Page 27). The distribution function of X is
.
We deduce that:
(2.11)
The joint distribution
of the random vector
is singular, whose support is the set
. Its distribution is:
on the set
.
2.3.4. Some Functions an Operators
Before moving on the next section, let’s introduce some functions and operators that will be used in this work.
The function
is defined by:
(2.12)
The Laplace transform of a function f is defined by:
(2.13)
The Dickson Hipp operator
and some of its properties are given by:
(2.14)
(2.15)
(2.16)
(2.16)
where
,
.
Many properties on Dickson Hipp operator can be consulted in [34] [35] [36] [37] .
3. Integro-Differential Equations Satisfied by the Gerber-Shiu Function
and
,
The main goal of this section is to show that the Gerber-Shiu function
and
satisfy some integro-differential equation. To achieve that, some notations and preliminaries are introduced.
3.1. Results and Preliminary
Let’s put:
a Brownian motion starting from zero with drift
and variance
;
is the running supremum of
;
is the first hitting time of the value
.
By Formula (2.0.2) of ( [34] , pp. 295), we have for
,
or
(3.1)
For
, we define the following potential measure
(3.2)
where:
is a Brownian motion starting from zero with drift
and variance
;
X is a random variable representing the amount of claim when a disaster occurs.
is a random variable representing the inter-claim time of exponential law with parameter
;
is an indicator function, which equals 1 if event A occurs and 0 otherwise;
u is the initial surplus with:
;
;
.
The potential measure
plays an important role in analyzing the Gerber-Shiu functions
and
.
In order to determine the potential measure in relation (3.2), let’s first calculate the following measure:
(3.3)
where:
is a random variable of exponential law with parameter q;
u is defined in relation (3.2), with
;
;
.
Lemma 3.1 ( [34] ). Assume that
is independent of
. Then the following variables
and
are independent and exponentially distributed with respective rates:
(3.4)
For
,
,
(3.5)
For
,
,
(3.6)
Relations (3.5) and (3.6) show that for
, the potential measure
is absolutely continuous w.r.t Lebesgue measure.
From (3.5) and (3.6), we can obtain the following result:
Lemma 3.2: For
, the measure
has a density given by
(3.7)
For
, the measure
has a density given by
(3.8)
where:
(3.9)
Proof. En conditioning on the value of
, we have:
(3.10)
By combining this last relation with relations (3.5) and (3.6), the expected result is obtained.
3.2. Integro-Differential Equations Satisfied by the Functions
,
Theorem 3.1. The Gerber-Shiu functions
in the risk model defined by relation (1.1) verify the following integro-diffferential equations.
(3.11)
(3.12)
where
(3.13)
(3.14)
(3.15)
(3.16)
(3.17)
(3.18)
(3.19)
(3.20)
(3.21)
(3.22)
Proof of relation (3.1)
By conditioning on the time and the amount of the first claim and taking into account the fact that ruin may occur or not, he follows:
(3.23)
By setting:
Relation (3.23) becomes:
(3.24)
Let’s calculate
,
(3.25)
By relation (3.13), relation (3.25) becomes:
(3.26)
By relation (3.14), relation (3.26) becomes:
(3.27)
By lemma 3.2, relation (3.27) can be put in the form:
(3.28)
Let’s calculate
and
.
The distribution
of the random vector
is a comonotonic distribution whose support is the set:
The distribution
is
on
.
Let’s calculate
,
(3.29)
where:
(3.30)
By injecting relation (3.30) into relation (3.29), he follows:
(3.31)
By lemma 3.2, relation (3.31) can be written:
(3.32)
Let’s calculate I4,1:
(3.33)
By lemma 3.2, relation (3.33) becomes:
(3.34)
Let’s calculate
.
Using relations (3.28) and (3.34), we have:
(3.35)
By Formulas (3.28) and (3.35) the function
can be put in the form:
(3.36)
where
and
are defined in relation (3.16).
A change of variable
, brings (3.36) into:
(3.37)
By relations (3.17) and (3.18), relation (3.37) can be put in the form:
(3.38)
Relation (3.38) can be written:
Proof of relation (3.12):
(3.39)
By setting:
Relation (3.39) becomes:
(3.40)
Let’s calculate
.
(3.41)
By relation (3.13), relation (3.41) can be written:
(3.42)
By relation (3.14), relation (3.42) becomes:
(3.43)
By lemma 3.2, relation (3.43) can be put in the form:
(3.44)
Let’s calculate
and
.
The distribution
random vector
is a comonotonic distribution whose support is the set:
.
The distribution
is
on
.
Let’s calculate
.
(3.45)
where:
(3.46)
By injecting relation (3.46) into relation (3.45), he follows:
(3.47)
By lemma 3.2, relation (3.47) can be written: