Applying Lundberg’s Ruin Bound

Why Apply a Ruin Bound?

In a surplus model, an insurer begins with capital, collects premium continuously, and pays claims as they occur. A central question is whether the surplus ever falls below zero. Exact ultimate ruin probabilities can be difficult to calculate, but the adjustment coefficient provides a useful upper bound when the classical model’s assumptions hold. The bound is not generally the exact probability: it is a conservative estimate derived from the model parameters.

This tutorial focuses on applying the bound. The key practical steps are to check that the model has positive net profit, find or verify the positive adjustment coefficient, use the initial capital in the exponent, and interpret the result as an upper bound on ultimate ruin.

Definition: In the classical compound Poisson surplus model, the surplus at time $t$ is $U(t)=u+ct-\sum_{j=1}^{N(t)}X_j$. Here $u\geq 0$ is initial capital, $c>0$ is the premium rate in dollars per unit time, $N(t)$ is a Poisson process with rate $\lambda>0$, and the claim amounts $X_1,X_2,\ldots$ are independent and identically distributed nonnegative random variables, independent of $N(t)$. The ultimate ruin probability is $\psi(u)=P(\text{there exists }t\geq 0\text{ such that }U(t)<0)$.

The model assumes a constant premium rate and Poisson claim arrivals, with independent claim amounts drawn from a fixed severity distribution. Let $M_X(r)=E[e^{rX}]$ be the moment-generating function of a claim amount, assumed finite at the positive values of $r$ used below. The net profit condition is $c>\lambda E[X]$: expected premium income per unit time exceeds expected claim outgo per unit time. Under the usual conditions for the classical model, this condition gives a positive adjustment coefficient.

Definition: The positive adjustment coefficient $R$ is a positive solution of $\lambda(M_X(R)-1)=cR$, with $M_X(R)$ finite. For $R>0$, Lundberg’s inequality states that $\psi(u)\leq e^{-Ru}$. The coefficient $R$ has units reciprocal to the units of claim amounts, so that $Ru$ is dimensionless.

For a model with annual claim frequency, $c$ is dollars per year and $\lambda$ is claims per year. Both sides of the adjustment equation then have units per year. The exponent $Ru$, in contrast, has no units. The claim amount and premium rate must use the same currency unit, and all rates must use the same time unit.

Why the Exponential Bound Holds

A short martingale argument explains why the adjustment equation produces an upper bound. Write $S(t)=\sum_{j=1}^{N(t)}X_j$. For a compound Poisson aggregate claim amount, independence of increments gives $E[e^{R S(t)}]=\exp\{\lambda t(M_X(R)-1)\}$. Therefore,

E[e^{-R U(t)}] =e^{-Ru-Rct}E[e^{R S(t)}] =e^{-Ru}\exp\{t[\lambda(M_X(R)-1)-cR]\}.

When $R$ satisfies the adjustment equation, the time-dependent factor is one. More strongly, the independent-increment model makes $e^{-R U(t)}$ a nonnegative martingale. Let $\tau$ be the first time the surplus is strictly negative. For any fixed $T$, optional stopping at the bounded time $\tau\wedge T$ gives $E[e^{-R U(\tau\wedge T)}]=e^{-Ru}$. On the event $\tau\leq T$, $U(\tau)<0$, and hence $e^{-R U(\tau)}>1$. Since the stopped quantity is nonnegative,

P(\tau\leq T)\leq E[e^{-R U(\tau\wedge T)}]=e^{-Ru}.

As $T$ increases, the events $\{\tau\leq T\}$ increase to the event of ultimate ruin. Taking the limit yields $\psi(u)\leq e^{-Ru}$. This argument explains the role of the positive root: it makes the exponential process a martingale. The zero root also solves the adjustment equation, but it gives only the uninformative bound $e^0=1$.

Applying the Bound in Practice

Check the model and net profit condition.
Confirm the stated Poisson frequency, claim severity, and premium rate use consistent time and currency units. Check $c>\lambda E[X]$.

Find the positive adjustment coefficient.
Substitute the severity moment-generating function into $\lambda(M_X(R)-1)=cR$. Distinguish the positive root from the ever-present zero root.

Check the coefficient and calculate the bound.
Verify both sides of the adjustment equation, then calculate $e^{-Ru}$ using the specified initial capital.

State what the result means.
Report the value as an upper bound on ultimate ruin probability under the model assumptions, not as an exact probability or a finite-time probability.

Worked Example: Exponential Claims

A hypothetical insurer has Poisson claim arrivals at rate $\lambda=2$ per year. Claim amounts are exponential with mean $\$1{,}000$, and the premium rate is $c=\$2{,}500$ per year. Initial capital is $u=\$5{,}000$. Calculate the Lundberg upper bound on ultimate ruin probability.

Method and model check. For exponential claims with mean $\mu=\$1{,}000$, $M_X(r)=1/(1-1000r)$, for $r<1/1000$ per dollar. The expected claim outgo per year is $2(1000)=\$2{,}000$, so $c=2500>2000$: the net profit condition holds.

Find the positive root. The adjustment equation is $2(1/(1-1000R)-1)=2500R$. Since $1/(1-1000R)-1=1000R/(1-1000R)$, this becomes $2000R/(1-1000R)=2500R$. The root $R=0$ is not the positive adjustment coefficient. Dividing by $R>0$ gives $2000/(1-1000R)=2500$, so $1-1000R=2000/2500=0.8$, and $R=0.0002$ per dollar.

Check and apply. At this value, $M_X(R)=1/0.8=1.25$. The left side of the adjustment equation is $2(1.25-1)=0.5$ per year, and the right side is $2500(0.0002)=0.5$ per year. Thus the root checks. The exponent is $Ru=0.0002(5000)=1$, so the bound is $e^{-1}=0.3679$, to four decimal places. Under this model, ultimate ruin probability is at most $0.3679$.

Common mistake avoided: Treating the zero root as the adjustment coefficient would produce the bound 1, which is valid but uninformative. The calculation requires the positive root.

Worked Example: Fixed Claim Amount and a Premium-Rate Check

A hypothetical portfolio has Poisson claim arrivals at rate $2$ per year, and every claim is exactly $\$1{,}000$. The premium rate is $c=\$2{,}000/\ln 2$ per year, and initial capital is $\$5{,}000$. Verify the proposed adjustment coefficient $R=\ln 2/1{,}000$ per dollar, then calculate the bound.

Method and model check. For a fixed claim of $\$1{,}000$, $M_X(R)=e^{1000R}$. Expected claim outgo is $2(1000)=\$2{,}000$ per year. Since $\ln 2<1$, $c=2000/\ln 2>2000$, so the net profit condition holds.

Verify the coefficient. At $R=\ln 2/1000$, $e^{1000R}=e^{\ln 2}=2$. The left side of the adjustment equation is $2(2-1)=2$ per year. The right side is $(2000/\ln 2)(\ln 2/1000)=2$ per year. Both sides agree, and $R>0$, so the proposed value is the positive adjustment coefficient.

Calculate the bound. The exponent is $Ru=(\ln 2/1000)(5000)=5\ln 2$. Therefore $e^{-Ru}=e^{-5\ln 2}=2^{-5}=1/32=0.03125$, or $0.0313$ to four decimal places. The model’s ultimate ruin probability is at most $0.0313$.

Common mistake avoided: The premium rate must be used exactly as stated. If it were mistakenly written as $2{,}000{,}000/\ln 2$, the right side at the proposed $R$ would be $2{,}000$, not $2$, and that $R$ would not solve the adjustment equation. Checking both sides catches the mismatch before applying the bound.

Worked Example: Two-Point Claim Severity

A hypothetical insurer has Poisson claim arrivals at rate $1$ per year. Each claim is $\$500$ or $\$1{,}000$, each with probability $1/2$. The premium rate is $c=\$1{,}000/\ln 2$ per year, and initial capital is $\$2{,}000$. Verify $R=\ln 2/500$ per dollar and calculate the bound.

Method and model check. The mean claim amount is $(500+1000)/2=\$750$, so expected claim outgo is $\$750$ per year. Since $\ln 2<1$, the premium rate $1000/\ln 2$ exceeds $\$1{,}000$ per year and thus exceeds expected claim outgo.

Verify the coefficient. For this two-point severity, $M_X(R)=(e^{500R}+e^{1000R})/2$. At $R=\ln 2/500$, the two exponentials are $e^{\ln 2}=2$ and $e^{2\ln 2}=4$, so $M_X(R)=(2+4)/2=3$. The left side is $1(3-1)=2$ per year. The right side is $(1000/\ln 2)(\ln 2/500)=2$ per year. The proposed positive coefficient therefore satisfies the equation.

Calculate the bound. Here $Ru=(\ln 2/500)(2000)=4\ln 2$. Thus $e^{-Ru}=2^{-4}=1/16=0.0625$. The ultimate ruin probability is at most $0.0625$.

Common mistake avoided: For a discrete severity, the moment-generating function is the probability-weighted average of $e^{RX}$. It is not $e^{R E[X]}$; those expressions generally differ.

Interpreting Results and Avoiding Traps

A smaller bound indicates a stronger guarantee from this inequality, but it does not establish the exact ruin probability. In particular, an upper bound of $0.0625$ does not imply that ruin occurs with probability $0.0625$. The bound may be conservative. It is for ultimate ruin, which considers all future time; it is not a calculation of ruin by a specified date.

Before reporting a result, check the inputs and interpretation. The premium rate $c$ is total premium income per unit time, not the premium per policy unless the model has only one policy. The frequency $\lambda$ and $c$ must use the same time unit. The severity distribution must be used in its stated currency unit when solving for $R$. Finally, the net profit condition and a finite moment-generating function at $R$ are essential model checks; a numerical root should not be applied mechanically when these conditions fail.

Check	What to verify	Why it matters
Net profit	$c>\lambda E[X]$	A positive adjustment coefficient for the classical model relies on positive expected drift.
Positive root	$R>0$, not $R=0$	The zero root gives the trivial bound 1.
Equation	$\lambda(M_X(R)-1)=cR$	A proposed coefficient must match the premium rate actually given.
Units	$Ru$ is dimensionless	Mixing dollars and thousands of dollars changes the result.
Interpretation	State an upper bound on ultimate ruin	The inequality is not an exact or finite-horizon probability.

Key takeaway: For a valid classical surplus model, verify the positive adjustment coefficient in the equation $\lambda(M_X(R)-1)=cR$, then calculate $e^{-Ru}$. The result is an upper bound on ultimate ruin probability, and its reliability depends on consistent units and the model assumptions.

Check Your Understanding

For each problem, check the net profit condition and the adjustment equation before applying the bound. Report probabilities to four decimal places where needed.

A hypothetical insurer has exponential claim amounts with mean $\$500$, Poisson claim frequency $3$ per year, premium rate $\$1{,}800$ per year, and initial capital $\$6{,}000$. Find the positive adjustment coefficient and the Lundberg bound. (Answer: $R=1/3{,}000$ per dollar; bound $e^{-2}=0.1353$.)
Claims arrive at rate $2$ per year, and each claim is exactly $\$400$. Premium is $\$800/\ln 2$ per year, and initial capital is $\$1{,}600$. Verify $R=\ln 2/400$ per dollar and calculate the bound. (Answer: both sides of the adjustment equation equal $2$ per year; bound $1/16=0.0625$.)
Claims arrive at rate $2$ per year and are $\$250$ or $\$750$, each with probability $1/2$. Premium is $\$2{,}000/\ln 2$ per year, and initial capital is $\$1{,}000$. Verify $R=\ln 2/250$ per dollar and calculate the bound. (Answer: $M_X(R)=5$, both sides equal $8$ per year, and the bound is $1/16=0.0625$.)
A model has $c=\$900$ per year, $\lambda=2$ claims per year, and $E[X]=\$500$. Can the positive adjustment coefficient and the bound be applied under the standard net profit condition? Explain. (Answer: No; $\lambda E[X]=\$1{,}000$ per year exceeds $c=\$900$ per year, so the net profit condition fails.)

Check	What to verify	Why it matters
Net profit	\(c>\lambda E[X]\)	A positive adjustment coefficient for the classical model relies on positive expected drift.
Positive root	\(R>0\), not \(R=0\)	The zero root gives the trivial bound 1.
Equation	\(\lambda(M_X(R)-1)=cR\)	A proposed coefficient must match the premium rate actually given.
Units	\(Ru\) is dimensionless	Mixing dollars and thousands of dollars changes the result.
Interpretation	State an upper bound on ultimate ruin	The inequality is not an exact or finite-horizon probability.

What You'll Learn