The Renewal Equation for Ultimate Ruin

Why Condition on the First Claim?

Ultimate ruin is the probability that an insurer’s surplus eventually becomes negative. Conditioning on the first claim separates two possibilities: that claim causes ruin immediately, or it leaves a nonnegative surplus from which the process continues. The continuation probability is again an ultimate-ruin probability, now at the surplus left after the first claim. This self-similarity is the basis of a renewal equation.

Use the classical compound Poisson surplus model. Initial surplus is \(u\ge 0\), premiums arrive continuously at rate \(c>0\), and claim arrivals form a Poisson process with rate \(\lambda>0\). Claim amounts \(X_1,X_2,\ldots\) are independent and identically distributed nonnegative random variables, independent of the arrival process. Write \(X\) for a generic claim amount, \(F_X(x)=P(X\le x)\), and \(S_X(x)=P(X>x)\). Define ruin as the surplus becoming strictly negative.

Definition: The surplus at time \(t\) is \(U(t)=u+ct-\sum_{j=1}^{N(t)}X_j\), where \(N(t)\) is the number of claims by time \(t\). The ultimate-ruin probability from initial surplus \(u\) is \(\psi(u)=P(\text{there exists }t\ge 0\text{ such that }U(t)<0\mid U(0)=u)\).

The first interarrival time \(T\) has density \(\lambda e^{-\lambda t}\) for \(t\ge0\). Just before the first claim, the surplus is \(u+cT\). If that claim is \(x>u+cT\), ruin occurs. If \(x\le u+cT\), the surplus immediately after the claim is \(u+cT-x\); future arrivals and claim amounts are independent of what has happened so far, so the conditional probability of eventual ruin is \(\psi(u+cT-x)\).

The First-Claim Renewal Equation

Condition first on \(T=t\), then on the claim amount. The claim either exceeds the available surplus or does not. Integrating over the density of \(T\) gives the equation in terms of the severity distribution:

\psi(u)=\int_0^\infty \lambda e^{-\lambda t} \left[ S_X(u+ct)+\int_{[0,u+ct]}\psi(u+ct-x)\,dF_X(x) \right]dt.

The inner integral is a Stieltjes integral, so it applies to both discrete and continuous severities. Its interval includes the boundary \(x=u+ct\): a claim exactly equal to the available surplus leaves zero surplus, which is not ruin under the strict-negative convention. The immediate-ruin term uses \(S_X(y)=P(X>y)\), not \(P(X\ge y)\), for the same reason.

Changing variables from waiting time \(t\) to premium income \(s=ct\) gives an equivalent form:

\psi(u)=\frac{\lambda}{c}\int_0^\infty e^{-\lambda s/c} \left[ S_X(u+s)+\int_{[0,u+s]}\psi(u+s-x)\,dF_X(x) \right]ds.

The equation is an identity under the stated model; the net profit condition is not needed simply to condition on the first claim. For the usual nontrivial long-run insurance setting, one also assumes \(E[X]<\infty\) and \(c>\lambda E[X]\). The net profit condition ensures a positive expected drift of surplus and is central to many results about ultimate ruin. It does not turn the renewal equation into a closed-form solution: in general, the continuation term still involves the unknown function \(\psi\).

Build the pre-claim surplus.
At first-claim time \(t\), it is \(u+ct\).

Separate the outcomes.
A claim \(x>u+ct\) causes immediate ruin; a claim \(x\le u+ct\) leaves surplus \(u+ct-x\).

Average over the first claim.
Use \(S_X(u+ct)\) for immediate ruin and integrate \(\psi(u+ct-x)\) against \(dF_X(x)\) for continuation.

Average over the waiting time.
Weight the conditional probability by \(\lambda e^{-\lambda t}\) and integrate over \(t\ge0\).

Worked Example: A Two-Point Severity and a Boundary Check

A hypothetical insurer has initial surplus \(u=2\) (in thousands of dollars), premium rate \(c=1\) thousand dollars per year, and Poisson claim rate \(\lambda=0.5\) per year. Each claim is 1 or 3 thousand dollars, each with probability \(1/2\). Write the first-claim equation for \(\psi(2)\), and calculate the probability that the first claim causes ruin.

Method. Condition on the first arrival time \(t\). Immediately before that claim, surplus is \(2+t\). For \(0\le t<1\), a claim of 3 exceeds the surplus; a claim of 1 leaves \(1+t\). For \(t\ge1\), both possible claims are at most the surplus. Their post-claim surpluses are \(1+t\) after a claim of 1 and \(t-1\) after a claim of 3.

Since the first-arrival density is \(0.5e^{-0.5t}\), splitting at \(t=1\) gives

\begin{aligned} \psi(2) ={}&\int_0^1 0.5e^{-0.5t} \left[\frac12\psi(1+t)+\frac12\right]dt\\ &+\int_1^\infty 0.5e^{-0.5t} \left[\frac12\psi(1+t)+\frac12\psi(t-1)\right]dt. \end{aligned}

In the first integral, the term \(1/2\) is the probability-weighted immediate-ruin outcome conditional on the arrival time: the 3-thousand claim occurs with probability \(1/2\). The overall probability of first-claim ruin is therefore

\int_0^1 0.5e^{-0.5t}\left(\frac12\right)dt =\frac14\int_0^1e^{-0.5t}dt =\frac12(1-e^{-0.5}) =0.1967.

The remaining terms in the renewal equation describe continuation and are not part of the first-claim-ruin probability. This example also checks the boundary: at \(t=1\), a claim of 3 leaves zero, not negative, surplus. A single arrival time has probability zero under the continuous waiting-time distribution, but the boundary convention is still important when writing the equation.

Using a Continuous Severity Distribution

For a continuous claim distribution with density \(f_X\), the Stieltjes integral can be written as an ordinary integral. The first-claim equation then separates the severity tail from a weighted average of continuation probabilities:

\psi(u)=\int_0^\infty \lambda e^{-\lambda t} \left[ S_X(u+ct)+\int_0^{u+ct}\psi(u+ct-x)f_X(x)\,dx \right]dt.

The first term is the probability of immediate ruin given the pre-claim surplus, averaged over the first arrival time. The second integrates over all claim amounts that do not cause immediate ruin. If the severity distribution has atoms, replacing its Stieltjes integral with an integral against a density would omit those atoms and may give a wrong equation.

Worked Example: Uniform Claim Amounts

A hypothetical portfolio has \(u=1\) thousand dollars of initial surplus, \(c=1\) thousand dollars per year, and \(\lambda=0.4\) claims per year. Claim amounts are uniformly distributed from 0 to 2 thousand dollars, so \(f_X(x)=1/2\) on \([0,2]\). Write the renewal equation and calculate the probability that the first claim causes ruin.

At first-claim time \(t\), available surplus is \(1+t\). Immediate ruin is possible only when \(1+t<2\), or \(0\le t<1\). In that interval the severity tail is \(S_X(1+t)=[2-(1+t)]/2=(1-t)/2\). For \(t\ge1\), the tail is zero. The continuation integral includes claim amounts from zero up to the lesser of \(1+t\) and 2. Thus

\psi(1)=0.4\int_0^\infty e^{-0.4t} \left[ \frac{(1-t)_+}{2} +\frac12\int_0^{\min(1+t,\,2)}\psi(1+t-x)\,dx \right]dt,

where \((1-t)_+=\max(1-t,0)\). The immediate first-claim-ruin probability is

\begin{aligned} 0.4\int_0^1 e^{-0.4t}\frac{1-t}{2}\,dt &=0.2\frac{0.4-1+e^{-0.4}}{(0.4)^2}\\ &=0.0879. \end{aligned}

For the integral, \(\int_0^1e^{-at}(1-t)\,dt=(a-1+e^{-a})/a^2\); substituting \(a=0.4\) gives \((0.4-1+e^{-0.4})/0.16\), approximately \(0.4395\), and multiplying by \(0.2\) gives \(0.0879\). The continuation term remains in the equation because calculating the probability of ruin at the first claim is not the same as calculating ultimate ruin. Probabilities here are rounded to four decimal places.

The Integrated-Tail Renewal Form

Under the net profit condition, an often useful alternative expresses the severity distribution through its integrated tail. This form requires the additional assumption \(0<E[X]<\infty\). Define the safety loading ratio \(\rho=\lambda E[X]/c\), which then satisfies \(0<\rho<1\), and define the integrated-tail density \(g\) and survival function \(\bar G\) by

g(y)=\frac{S_X(y)}{E[X]}\quad(y\ge0), \qquad \bar G(u)=\int_u^\infty g(y)\,dy.

This is a probability density because, for a nonnegative severity, \(\int_0^\infty S_X(y)\,dy=E[X]\). In this notation, the classical ultimate-ruin renewal equation is

\psi(u)=\rho\,\bar G(u)+\rho\int_0^u\psi(u-y)g(y)\,dy.

The first term is a scaled integrated-tail survival probability; the convolution term accounts for continuation through the integrated-tail distribution. This is a useful form for renewal methods and numerical calculation. It is distinct from the first-arrival-time equation: the two describe the same ultimate-ruin probability, but condition and integrate in different ways.

The positive-mean assumption matters. If \(E[X]=0\) for a nonnegative claim amount, then \(X=0\) almost surely. In that degenerate case there are no positive losses and \(\psi(u)=0\) for \(u\ge0\); the definition of \(g\) would divide by zero and should not be used. A zero-mean severity is not a case with a positive integrated-tail density.

Worked Example: Constructing the Integrated-Tail Equation

Suppose a hypothetical insurer has claim rate \(\lambda=0.3\) per year, premium rate \(c=1.2\) thousand dollars per year, and claim amounts of 1 or 3 thousand dollars, with probabilities \(3/4\) and \(1/4\), respectively. Construct \(g\), \(\bar G(2)\), and the integrated-tail equation at \(u=2\).

First calculate the mean claim amount: \(E[X]=(3/4)(1)+(1/4)(3)=1.5\) thousand dollars. The expected claim cost per year is \(\lambda E[X]=(0.3)(1.5)=0.45\) thousand dollars, which is less than \(c=1.2\). Thus the net profit condition holds. Also \(E[X]>0\), as required for this formulation, and

\rho=\frac{0.45}{1.2}=0.375.

The severity survival function is 1 for \(0\le y<1\), \(1/4\) for \(1\le y<3\), and 0 for \(y\ge3\). Values at isolated endpoints do not change the density calculations. Therefore \(g(y)=2/3\) on \(0\le y<1\), \(g(y)=1/6\) on \(1\le y<3\), and zero thereafter. Its total integral is \((2/3)(1)+(1/6)(2)=1\), as required for a density. From \(u=2\) to infinity, only the interval from 2 to 3 contributes, so \(\bar G(2)=(1/6)(1)=1/6\).

Substitution into the renewal equation gives

\psi(2)=0.375\left[ \frac16+ \frac23\int_0^1\psi(2-y)\,dy+ \frac16\int_1^2\psi(2-y)\,dy \right].

The convolution only integrates over \(0\le y\le u=2\); the part of \(g\) above 2 contributes to \(\bar G(2)\), not to that convolution. This example constructs the renewal equation and its coefficients; it does not claim to solve the resulting equation for \(\psi(2)\).

Common Mistakes and Why the Equation Matters

The renewal equation is useful because it turns an ultimate-time event into a relationship among ruin probabilities at different reserve levels. It also makes clear which parts depend on claim frequency, premium income, and severity. The formula is a starting point for analytic or numerical methods, not a guarantee that a closed-form answer exists.

Using the wrong post-claim reserve. If the pre-claim reserve is \(u+ct\) and claim amount is \(x\), the reserve is \(u+ct-x\). For example, with \(u=2,c=1,x=3\), after \(t\ge1\) the reserve is \(t-1\), not \(3+t\).
Mixing up strict and non-strict ruin. Under the convention here, \(x=u+ct\) leaves zero and is not immediate ruin. Use \(S_X(y)=P(X>y)\) for this convention. A different definition of ruin at zero requires corresponding boundary changes.
Dropping continuation. The probability that the first claim causes ruin is only the \(S_X\) component. Ultimate ruin also includes paths on which the first claim is survived but a later claim causes ruin.
Assuming every severity has a density. Discrete and mixed distributions require integration with respect to \(dF_X\); a density-only expression can lose point masses.
Using the integrated-tail form at zero mean. The formula for \(g\) requires \(0<E[X]<\infty\). If the nonnegative claims are zero almost surely, handle that degenerate case directly.

The first-claim equation itself has no interest-rate or mortality assumption: it is a surplus model with continuous premiums and Poisson claims. Its essential modeling choices are the arrival process, claim-severity law, premium rate, and convention for ruin. Making those choices explicit is necessary before applying or comparing any renewal formulation.

Check Your Understanding

For each problem, identify the correct conditioning or renewal terms and check the stated ruin convention.

A hypothetical insurer has \(u=4\), \(c=2\), and \(\lambda=0.5\), in consistent monetary and time units. Write the first-claim equation using \(F_X\) and \(S_X\). (Answer: \(\displaystyle \psi(4)=\int_0^\infty0.5e^{-0.5t}\left[S_X(4+2t)+\int_{[0,4+2t]}\psi(4+2t-x)\,dF_X(x)\right]dt\).)
At a first arrival time \(t=2\), an insurer with \(u=1\) and \(c=1\) has pre-claim surplus 3. A claim of 4 occurs. What happens immediately, and what is the post-claim surplus? (Answer: Immediate ruin; the post-claim surplus is \(-1\).)
Claims are 2 or 5 with equal probability. For pre-claim surplus \(y=3\), what are the immediate-ruin probability and the continuation contribution conditional on that surplus? (Answer: Immediate-ruin probability \(1/2\); continuation contribution \(\frac12\psi(1)\).)
A severity is uniform on \([0,4]\), with \(u=2\), \(c=1\), and \(\lambda=0.25\). Write the immediate-ruin component of the first-claim equation only. (Answer: \(\displaystyle 0.25\int_0^2 e^{-0.25t}\frac{2-t}{4}\,dt\); for \(t\ge2\), the immediate-ruin tail is zero.)
Claims are 1 or 3 with probabilities \(3/4\) and \(1/4\), and \(c=1.5,\lambda=0.4\). Calculate \(E[X]\), check the net profit condition, and find \(\rho\). (Answer: \(E[X]=1.5\), \(\lambda E[X]=0.6<1.5=c\), and \(\rho=0.6/1.5=0.4\).)

What You'll Learn