Integrated Classical Ruin Model Problems

Integrating the Classical Ruin Model

A ruin calculation is reliable only when its model, time scale, and event definition agree. In the classical model, premium income accumulates continuously, while claims arrive at random times and reduce surplus by their full amounts. This tutorial brings those pieces together: first specify the model, then track a path or calculate a ruin probability, and finally check what an ultimate ruin bound does—and does not—say.

Definition: In the classical compound Poisson surplus model, initial surplus is $u\geq 0$, premium income arrives at rate $c>0$, claims arrive according to a Poisson process $N(t)$ with rate $\lambda>0$, and claim amounts $X_1,X_2,\ldots$ are independent and identically distributed, independent of $N(t)$. The surplus at time $t$ is $$ U(t)=u+ct-\sum_{j=1}^{N(t)}X_j. $$ Ruin time is $\tau=\inf\{t\geq 0:U(t)<0\}$, and the ultimate ruin probability is $\psi(u)=P(\tau<\infty)$. Ruin here means strictly negative surplus; reaching zero alone is not ruin. The net profit condition is $c>\lambda E[X]$.

Rates and times must use matching units. For example, if $\lambda$ is claims per year, $c$ must be dollars per year and all times in $U(t)$ must be in years. A common premium convention is $c=(1+\theta)\lambda E[X]$, where $\theta>0$ is the safety loading. State this convention when using it; the net profit condition itself only requires $c>\lambda E[X]$.

Read the Surplus Path Before Calculating a Probability

Between claims, surplus increases linearly with slope $c$. At a claim time, it falls by that claim amount. Since claims arrive at isolated times in this model, a path can be checked by calculating surplus immediately before and after each claim. If it becomes negative just after a claim, ruin has occurred, and subsequent premiums cannot reverse that event.

Worked Example: A Discrete-Claim Surplus Path

A hypothetical insurer starts with $u=\$2$. Claims arrive at rate $\lambda=1$ per year. Claim amounts are $\$1$ with probability $0.75$ and $\$3$ with probability $0.25$. The insurer uses safety loading $\theta=0.20$. On one observed path, a $\$3$ claim arrives at $t=0.5$ years. Determine whether ruin occurs on that path.

Method and model check. The mean claim amount is $E[X]=0.75(1)+0.25(3)=1.50$ dollars. Thus expected claims per year are $\lambda E[X]=\$1.50$, and the premium rate under the stated loading convention is $c=(1+0.20)(1)(1.50)=\$1.80$ per year. The net profit condition holds because $1.80>1.50$.

Track the path. Before the claim at $t=0.5$, the surplus is $U(0.5-)=2+1.80(0.5)=\$2.90$. Immediately after paying the $\$3$ claim, it is $U(0.5)=2.90-3=-\$0.10$.

Because the post-claim surplus is strictly negative, ruin occurs at $t=0.5$ years. The calculation is internally consistent: $\$1.80$ per year multiplied by $0.5$ years gives $\$0.90$ of premium income. Common mistake avoided: comparing the claim with initial surplus alone would ignore the premium received before its arrival.

Adjustment Coefficients and the Ultimate Bound

The adjustment coefficient summarizes how the claim distribution and premium rate jointly affect the exponential bound on ultimate ruin. For claim moment-generating function $M_X(r)=E[e^{rX}]$, a positive adjustment coefficient $R$ solves the Lundberg equation below. The moment-generating function must be finite at $R$, and the model must satisfy the net profit condition. When these conditions hold, the standard Lundberg bound is $\psi(u)\leq e^{-Ru}$.

\lambda\bigl(M_X(R)-1\bigr)=cR,\qquad R>0, \qquad\text{and}\qquad \psi(u)\leq e^{-Ru}.

The equation compares the exponential growth generated by claim arrivals with the premium drift. It is not a claim that the bound is the exact ruin probability. In some severity models the adjustment equation has a convenient closed-form solution; in others the positive root must be found numerically. A proposed root should always be substituted back into the equation.

Derivation: Define $S(t)=\sum_{j=1}^{N(t)}X_j$. For an adjustment coefficient $R$, the compound Poisson moment-generating function gives $E[e^{R S(t)}]=\exp\{\lambda t(M_X(R)-1)\}$. Therefore $$ E[e^{-R U(t)}] =e^{-Ru}\exp\{t[\lambda(M_X(R)-1)-cR]\} =e^{-Ru} $$ when $R$ solves the Lundberg equation. The corresponding process $e^{-R U(t)}$ is a nonnegative martingale. At ruin, $U(\tau)<0$, so $e^{-R U(\tau)}>1$. Applying the martingale bound at ruin truncated at time $n$ gives $P(\tau\leq n)\leq e^{-Ru}$; letting $n$ increase yields $\psi(u)\leq e^{-Ru}$. This is an upper bound, not an equality in general.

Worked Example: Exponential Claims and an Ultimate Ruin Bound

A hypothetical portfolio has claim arrival rate $\lambda=0.4$ per year and claim amounts $X\sim\operatorname{Exponential}(\beta=0.5)$, where $\beta$ is the rate in inverse dollars. The insurer charges a safety loading $\theta=0.25$, and initial surplus is $u=\$4$. Find the premium rate, adjustment coefficient, Lundberg bound, and exact ultimate ruin probability.

Method and assumptions. The exponential mean is $E[X]=1/\beta=2$ dollars, so the expected claim cost per year is $0.4(2)=\$0.80$. The premium rate is $c=(1+0.25)(0.4)(2)=\$1.00$ per year; hence $c>\lambda E[X]$. For $r<\beta$, $M_X(r)=\beta/(\beta-r)$. Substituting this into the adjustment equation gives $0.4(0.5/(0.5-R)-1)=R$.

Solve and verify. Since $0.5/(0.5-R)-1=R/(0.5-R)$, the equation for a positive root reduces to $0.4R/(0.5-R)=R$. Dividing by $R>0$ gives $0.4=0.5-R$, so $R=0.10$ per dollar. Substitution verifies the result: the left side is $0.4(0.5/0.4-1)=0.4(0.25)=0.10$, and the right side is $1.00(0.10)=0.10$.

Calculate the bound. The Lundberg bound is $\psi(4)\leq e^{-0.10(4)}=e^{-0.4}=0.6703$, to four decimal places.

Check against the exponential-claim exact result. In this classical model with exponential claims, the exact ultimate ruin probability is $\psi(u)=[\lambda/(c\beta)]e^{-Ru}$. Thus $\psi(4)=[0.4/(1.00(0.5))]e^{-0.4}=0.8(0.670320\ldots)=0.5363$, to four decimal places. This is below the bound, as required. Common mistake avoided: treating $e^{-Ru}$ as the exact probability would give $0.6703$ rather than $0.5363$.

Discrete Severity: Verify the Root, Then Use It

For discrete claim sizes, the moment-generating function is a probability-weighted sum of exponentials. The adjustment equation still applies, but a root may not simplify unless the model parameters are chosen accordingly. A useful exam check is to verify both the premium calculation and the candidate root, rather than relying on a rounded numerical solver output.

Worked Example: A Designed Adjustment Root for Discrete Claims

A hypothetical insurer has $\lambda=1$ claim per year and $u=\$3$. Each claim is $\$1$ or $\$2$, each with probability $0.5$. The annual premium rate is $c=2/\ln 2$ dollars per year. Find the safety loading, adjustment coefficient, and ultimate ruin bound.

Calculate the loading. The mean claim amount is $E[X]=0.5(1)+0.5(2)=\$1.50$, so expected claim cost is $\lambda E[X]=\$1.50$ per year. From $c=(1+\theta)\lambda E[X]$, $$ \theta=\frac{c}{\lambda E[X]}-1 =\frac{2/\ln 2}{1.5}-1 =\frac{4}{3\ln 2}-1 \approx 0.9236. $$ The net profit condition holds since $2/\ln 2\approx2.8854>1.50$.

Verify the adjustment coefficient. Here $M_X(r)=0.5e^r+0.5e^{2r}$. At $R=\ln 2$, $e^R=2$ and $e^{2R}=4$, so $M_X(R)=0.5(2)+0.5(4)=3$. The two sides of the Lundberg equation are $\lambda(M_X(R)-1)=1(3-1)=2$ and $cR=(2/\ln 2)(\ln 2)=2$. Thus $R=\ln 2$ is a positive root.

Apply the bound. With $u=3$, $\psi(3)\leq e^{-3\ln 2}=2^{-3}=0.1250$. This is an upper bound on ultimate ruin, not an exact probability. Common mistake avoided: inserting $e^R$ and $e^{2R}$ as if they were the claim amounts; the moment-generating function uses exponentials of the amounts.

Finite-Horizon Ruin and Claim-Count Decomposition

A finite-horizon event specifies both a time interval and a ruin condition. Conditional on one claim arriving at time $s$, ruin at that claim occurs if its amount exceeds the surplus immediately before payment, $u+cs$. For a Poisson process, the probability density of exactly one arrival in $[0,T]$, at time $s$, is $e^{-\lambda T}\lambda\,ds$. Integrating the severity tail over possible arrival times therefore gives the probability of ruin caused by exactly one claim by time $T$:

P\bigl(N(T)=1,\ \text{ruin by }T\bigr) =e^{-\lambda T}\lambda\int_0^T S_X(u+cs)\,ds.

This is one claim-count contribution, not generally the entire finite-horizon ruin probability: paths with two or more claims may also ruin. The zero-claim contribution is zero when $u\geq0$ and $c>0$. For multiple claims, ruin must be checked at each claim time using the cumulative prior claims and premium income received by then.

Worked Example: Exactly One Claim Causes Ruin by a Fixed Time

A hypothetical insurer has $u=\$2$, claim arrival rate $\lambda=0.5$ per year, and $X\sim\operatorname{Exponential}(\beta=1)$, with amounts in dollars. It charges safety loading $\theta=0.5$ and is evaluated over $T=2$ years. Calculate the probability that exactly one claim arrives by year 2 and causes ruin.

Check the premium rate. Since $E[X]=1/\beta=\$1$, expected claim cost is $\lambda E[X]=\$0.50$ per year. The premium rate is $c=(1+0.5)(0.5)(1)=\$0.75$ per year, which satisfies the net profit condition.

Set up the event. If the single claim arrives at time $s$, the insurer has $2+0.75s$ dollars immediately before it. For an exponential claim, the probability that the claim exceeds this amount is $S_X(2+0.75s)=e^{-(2+0.75s)}$. The probability density contribution for exactly one arrival at $s$ is $e^{-0.5(2)}(0.5)\,ds$.

Integrate. $$ e^{-1}(0.5)\int_0^2 e^{-(2+0.75s)}\,ds =\frac{e^{-3}}{1.5}\bigl(1-e^{-1.5}\bigr) =0.0258. $$ The answer is $0.0258$ to four decimal places. The integral uses time in years throughout, and the rate $0.75$ is dollars per year. Common mistake avoided: using $P(X>u)$ for every possible arrival time ignores the premium accumulated before the claim. This result counts only paths with exactly one claim; it does not include ruin paths with two or more claims.

Integrated Checks and Common Traps

A sound solution usually follows the same sequence: specify the surplus model and units; calculate $E[X]$ and check $c>\lambda E[X]$; track claim times and amounts or identify the requested event; solve and verify $R$ if an adjustment bound is needed; and label the result as exact, conditional, or an upper bound. These labels matter. A probability for exactly one claim is not the probability of all finite-horizon ruin, and an ultimate bound is not a finite-horizon probability.

Mixing time units: if $\lambda$ is per year, use $t$ in years and $c$ per year.
Ignoring claim timing: premium earned before a claim depends on when that claim arrives.
Using a bound as an equality: $\psi(u)\leq e^{-Ru}$ does not generally determine $\psi(u)$ exactly.
Changing the ruin definition: with strict ruin, $U(t)=0$ is not ruin; a claim that takes surplus below zero is.
Overlooking omitted claim counts: the exactly-one-claim term excludes paths with multiple claims.

Key takeaway: Integrated ruin problems are controlled by consistent accounting. Premiums accumulate over elapsed time, claim amounts enter at their arrival times, and every probability must match its event: a particular path, a claim-count contribution, a finite-horizon event, or an ultimate bound.

Check Your Understanding

For each problem, keep money and time units consistent and identify whether the requested result is a path outcome, a claim-count contribution, or a bound.

A hypothetical insurer has initial surplus $\$3$, premium rate $\$1.20$ per year, and claims of $\$2$ at $t=0.5$ and $\$3$ at $t=1.25$ years. Find the surplus after each claim and the first ruin time, if any. (Answer: After the first claim surplus is $\$1.60$; immediately before the second it is $\$2.50$, and immediately after it is $-\$0.50$; ruin occurs at $1.25$ years.)
Claims arrive at $\lambda=0.3$ per year and have exponential rate $\beta=0.6$ per dollar. With safety loading $\theta=0.5$, find $c$, $R$, and the Lundberg bound at $u=\$5$. (Answer: $c=\$0.75$ per year, $R=0.20$ per dollar, and $\psi(5)\leq e^{-1}=0.3679$.)
Let $\lambda=1$ per year, $T=1$ year, $u=\$1$, $c=\$1$ per year, and $X\sim\operatorname{Exponential}(\beta=1)$. Find the probability of exactly one claim by $T$ that causes ruin. (Answer: $e^{-2}(1-e^{-1})=0.0855$, to four decimal places.)
Claims arrive at $\lambda=1$ per year and are independently $\$1$ or $\$3$, each with probability $0.5$. If $c=4/\ln 2$ dollars per year and $u=\$2$, verify $R=\ln 2$ and calculate the Lundberg bound. (Answer: $E[X]=2$, $M_X(\ln 2)=5$, and both sides of the adjustment equation equal $4$; $\psi(2)\leq e^{-2\ln 2}=0.2500$.)
For $\lambda=0.5$ per year, $X\sim\operatorname{Exponential}(\beta=1)$, $c=\$0.75$ per year, $u=\$2$, and $T=2$ years, distinguish the probability that exactly one claim causes ruin from the full finite-horizon ruin probability. (Answer: The exactly-one-claim contribution is $\frac{e^{-3}}{1.5}(1-e^{-1.5})=0.0258$; the full finite-horizon probability is at least $0.0258$, because it also includes possible ruin paths with two or more claims.)

What You'll Learn