Saddlepoint approximations to the probability of ruin in finite time for the compound Poisson risk process perturbed by diffusion (Complement)

Riccardo Gatto    Benjamin Baumgartner
2014

Introduction

This document serves as a complement to

[GB] Gatto, R. and Baumgartner, B. (2016). “Saddlepoint approximations to the probability of ruin in finite time for the compound Poisson risk process perturbed by diffusion”. Methodology and Computing in Applied Probability 18(1), pp. 217–235.

The joint or double Laplace transform of the time of ruin T and the initial capital Y0 is given in GB (Theorem 2.1). A proof using the Laplace transform of the Gerber–Shiu function is given in GB (Appendix) and a complete proof, which does not require the Gerber–Shiu function, is the following.

Alternative proof of Theorem 2.1

Within any small time interval of length h>0, either zero, one or more than one jumps (or claims) occur with the corresponding probabilities 1−λ⁢h+o⁡(h), λ⁢h+o⁡(h) and o⁡(h), as h↓0. Thus, from the independence and the stationarity of the increments of {Yt}t≥0, we obtain

fα⁢(x) =(1−λ⁢h)⁢𝔼⁢𝔼⁢[eα⁡(T+h)⁢𝟙⁢(T<∞)|Y0=x+c⁢h+σ⁢Wh]
+λ⁢h⁢𝔼⁢𝔼⁢[eα⁡(T+h)⁢𝟙⁢(T<∞)|Y0=x+c⁢h+σ⁢Wh−X1]+o⁡(h)
=(1−λ⁢h)⁢eα⁢h⁢𝔼⁢[fα⁢(x+c⁢h+σ⁢Wh)]+λ⁢h⁢eα⁢h⁢𝔼⁢[fα⁢(x+c⁢h+σ⁢Wh−X1)]+o⁡(h), (1)

as h↓0 and for any α∈ℝ such that fα⁢(x)<∞. The first expectation in (1) can be written as

𝔼⁡[fα⁢(x+c⁢h+σ⁢Wh)] =fα⁢(x)+𝔼⁡[∑k=12fα(k)⁢(x)k!⁢(c⁢h+σ⁢Wh)k+o⁡({c⁢h+σ⁢Wh}2)]
=fα⁢(x)+∑k=12(fα(k)⁢(x)k!⁢∑i=0k(ki)⁢(c⁢h)k−i⁢σi⁢𝔼⁢[Whi])+𝔼⁡[o⁡(Wh2)]
=fα⁢(x)+c⁢h⁢fα′⁢(x)+12⁢σ2⁢h⁢fα′′⁢(x)+o⁡(h). (2)

Similarly, the second expectation in (1) can be written as

𝔼⁡[fα⁢(x+c⁢h+σ⁢Wh−X1)] =𝔼⁡[fα⁢(x−X1)]+𝔼⁡[∑k=12fα(k)⁢(x−X1)k!⁢(c⁢h+σ⁢Wh)k+o⁡({c⁢h+σ⁢Wh}2)]
=𝔼⁡[fα⁢(x−X1)]+∑k=12(𝔼⁡[fα(k)⁢(x−X1)]k!⁢∑i=0k(ki)⁢(c⁢h)k−i⁢σi⁢𝔼⁢[Whi])+𝔼⁡[o⁡(Wh2)]
=𝔼⁡[fα⁢(x−X1)]+c⁢h⁢𝔼⁢[fα′⁢(x−X1)]+12⁢σ2⁢h⁢𝔼⁢[fα′′⁢(x−X1)]+o⁡(h). (3)

As indicated by a Referee, the existence of fα′′, required in the Taylor expansions in (2) and (3), is established by Feng (2011, Lemma C.1), because the function fα is a special case of the more general functional of T given in GB (Equation 15). Replacing the expectations in (1) with their respective expansions (2) and (3), dividing both sides by eα⁢h⁢h⁢(1−λ⁢h) and rearranging terms results in

0 =1h(1−e−α⁢h1−λ⁢h)fα(x)+cfα′(x)+12σ2fα′′(x)
+11−λ⁢h⁢(𝔼⁡[fα⁢(x−X1)]+c⁢h⁢𝔼⁢[fα′⁢(x−X1)]+12⁢σ2⁢h⁢𝔼⁢[fα′′⁢(x−X1)])+o⁡(1).

Let g⁡(h)=e−α⁢h⁢(1−λ⁢h)−1, then by the rule of de l’Hospital the coefficient of fα⁢(x) converges to limh↓0(g⁡(0)−g⁡(h))/h=−g′⁢(0)=α−λ. Thus, by letting h↓0, we obtain the integro-differential equation

0 =12⁢σ2⁢fα′′⁢(x)+c⁢fα′⁢(x)+(α−λ)⁢fα⁢(x)+λ⁢𝔼⁢[fα⁢(x−X1)]
or, equivalently,
0 =12⁢σ2⁢fα′′⁢(x)+c⁢fα′⁢(x)+(α−λ)⁢fα⁢(x)+λ⁢∫0xfα⁢(x−ξ)⁢d⁢FX⁢(ξ)+λ⁡[1−FX⁢(x)]. (4)

In the next step, both sides of (4) are multiplied by eβ⁢x and integrated from 0 to ∞. This corresponds to taking Laplace transforms with reversed sign of the argument β, thus fα′^⁢(β)=−β⁢f^α⁢(β)−fα⁢(0) and fα′′^⁢(β)=β2⁢f^α⁢(β)+β⁢fα⁢(0)−fα′⁢(0), for any β∈ℝ such that f^α⁢(β)<∞, where g^⁢(u)=∫0∞eu⁢x⁢g⁢(x)⁢𝑑x, for a generic function g. As a consequence,

0 =12⁢σ2⁢[β2⁢f^α⁢(β)+β⁢fα⁢(0)−fα′⁢(0)]+c⁡[−β⁢f^α⁢(β)−fα⁢(0)]
+(α−λ)⁢f^α⁢(β)+λ⁢f^α⁢(β)⁢MX⁢(β)−λβ⁢[1−MX⁢(β)],

which, when solved for f^α⁢(β), which is the left-hand side of (5), leads to

f^α⁢(β) =(c−12⁢σ2⁢β)⁢fα⁢(0)+12⁢σ2⁢fα′⁢(0)−λβ⁢(MX⁢(β)−1)12⁢σ2⁢β−c⁢β+α+λ⁡(MX⁢(β)−1)
=(c−12⁢σ2⁢β)⁢fα⁢(0)+12⁢σ2⁢fα′⁢(0)−κ⁡(β)β+12⁢σ2⁢β−cκ⁡(β)+α.

Note that f^α⁢(β) exists for all β<0, if α<0. In particular, it exists for β=v⁡(α)<0 with α<0. In this case the above denominator vanishes and therefore v⁡(α) is a common root of both the denominator and numerator above, otherwise the existence of f^α⁢(v⁢(α)) would be contradicted. Because of that, setting the numerator equal to 0 and substituting v⁡(α) for β yields

12⁢σ2⁢fα′⁢(0)=−(c−12⁢σ2⁢v⁢(α))⁢fα⁢(0)−αv⁡(α)−12⁢σ2⁢v⁢(α)+c,

and hence

f^α⁢(β)=12⁢σ2⁢(β−v⁡(α))⁢(1−fα⁢(0))−κ⁡(β)β−αv⁡(α)κ⁡(β)+α.

Without initial reserve, i. e. for x=0, ruin occurs almost surely and T=0 a. s. because the regularity of the Wiener process implies that the risk process crosses the null level infinitely often over any arbitrarily small time interval containing the origin. Hence fα⁢(0)=1 and, as a consequence, GB (Equation 12), i. e.

f^α⁢(β)=−αv⁡(α)+κ⁡(β)βα+κ⁡(β) (5)

holds for all α,β<0.

From the fact that D={(α,β)∈ℝ2:α≤α̊,β<v~(α)} is a connected subset of ℝ2 and the right-hand side of (5) is an analytical function for all α,β∈D, follows that the double Laplace transform formula (5) holds over the entire set D. ∎

A helpful picture of the domain D is given by Figure 1.

Figure 1: A representation of the domain D of the double Laplace transform f^α⁢(β)