The Net Profit Condition and Certain Ruin

Why Long-Run Drift Matters

In a classical surplus model, an insurer receives premium income continuously and pays claims as they occur. The balance between premium income and expected claim outgo determines whether surplus tends to rise or fall over the long run. This comparison is the net profit condition. It is a condition on the model’s average rates, not a promise about the path of any particular insurer: even positive average drift does not rule out a large claim and ruin.

The distinction between positive, zero, and negative drift is especially important for ultimate ruin. Negative drift leads to certain ruin under the assumptions below. With zero drift, certain ruin also holds for nondegenerate models, but its justification is different. One degenerate boundary case must be treated separately.

Definition: In the classical surplus model, the insurer starts with initial surplus $u\geq 0$, receives premium at constant rate $c\geq 0$, and pays claims generated by a Poisson process $N(t)$ with rate $\lambda>0$. Claim amounts $X_1,X_2,\ldots$ are independent and identically distributed, nonnegative, and independent of $N(t)$. Write $S(t)=\sum_{j=1}^{N(t)}X_j$ for aggregate claims and $U(t)=u+ct-S(t)$ for surplus at time $t$. Assume $E[X]$ is finite. Ruin occurs when surplus is strictly negative; the ruin time is $\tau_u=\inf\{t\geq 0:U(t)<0\}$, and the ultimate ruin probability is $\psi(u)=P(\tau_u<\infty)$.

The Net Profit Condition and Expected Drift

The expected number of claims by time $t$ is $\lambda t$. Since each claim has mean $E[X]$, expected aggregate claims are $\lambda t E[X]$. Therefore,

E[U(t)]=u+ct-\lambda tE[X] =u+(c-\lambda E[X])t.

This calculation identifies the mean surplus drift per unit time as $c-\lambda E[X]$. The net profit condition is $c>\lambda E[X]$: premium income exceeds expected claim outgo per unit time. The safety loading, when $E[X]>0$, is often expressed as $\theta=c/(\lambda E[X])-1$, so the net profit condition is equivalent to $\theta>0$. The drift comparison itself remains meaningful when $E[X]=0$, even though that ratio does not.

Key distinction: The net profit condition concerns the sign of the long-run average drift. If it holds, the model has positive drift, but ruin can still occur. If it fails, then under the nondegeneracy and moment assumptions stated below, ultimate ruin occurs with probability one. The zero-claim, zero-premium case is an exception.

Worked Example: Checking Positive Net Profit

A hypothetical insurer has claim frequency $\lambda=0.20$ claims per year and mean claim amount $E[X]=\$4{,}000$. Its premium income rate is $c=\$1{,}000$ per year, and its initial surplus is $u=\$10{,}000$. There is no investment income. Determine whether the net profit condition holds and calculate the expected surplus after five years.

Method. Compare premium income per year with expected claim outgo per year. Expected claim outgo is $\lambda E[X]=0.20(\$4{,}000)=\$800$ per year.

The drift is $c-\lambda E[X]=\$1{,}000-\$800=\$200$ per year. Since $\$1{,}000>\$800$, the net profit condition holds. After five years, $E[U(5)]=\$10{,}000+(\$200)(5)=\$11{,}000$.

Check. Directly, premium income over five years is $\$5{,}000$, and expected aggregate claims are $5(0.20)(\$4{,}000)=\$4{,}000$. Thus expected surplus is $\$10{,}000+\$5{,}000-\$4{,}000=\$11{,}000$, as above. This positive expected drift does not mean ruin is impossible. The common error is to treat an increasing expected surplus as a guarantee that every surplus path remains nonnegative.

Why Failure of the Condition Leads to Certain Ruin

For negative drift, the strong law of large numbers gives a direct argument. Let $T_n$ be the time of the $n$th claim and $W_j=T_j-T_{j-1}$ the interarrival times, with $T_0=0$. For a Poisson process, the $W_j$ are independent exponential random variables with mean $1/\lambda$. Consequently $T_n/n$ tends to $1/\lambda$ almost surely. The strong law also gives $\sum_{j=1}^n X_j/n\to E[X]$ almost surely. At the $n$th claim time,

\frac{U(T_n)}{n} =\frac{u}{n} +c\frac{T_n}{n} -\frac{1}{n}\sum_{j=1}^nX_j \longrightarrow \frac{c}{\lambda}-E[X].

If $c<\lambda E[X]$, this limit is strictly negative. Hence $U(T_n)$ is eventually negative almost surely. In particular, ruin has occurred by one of these claim times, so $\psi(u)=1$ for every finite $u$. This conclusion does not specify a fixed time by which ruin must occur.

At equality, $c=\lambda E[X]$, the limit above is zero and does not prove ruin: convergence to zero after division by $n$ does not tell us whether the unscaled surplus ever goes below zero. Instead, examine the surplus changes between successive claim times. Define $Z_j=cW_j-X_j$. Then $U(T_n)=u+\sum_{j=1}^n Z_j$, and the $Z_j$ are independent and identically distributed with $E[Z_j]=c/\lambda-E[X]=0$.

For a nondegenerate model with $c>0$ and finite claim-size variance, the increments have finite, strictly positive variance. Indeed, $W_j$ is independent of $X_j$, and $\operatorname{Var}(Z_j)=c^2/\lambda^2+\operatorname{Var}(X)$. The law of the iterated logarithm for independent, identically distributed, mean-zero increments with finite positive variance states that their partial sums have arbitrarily large negative values almost surely. In particular, for some claim epoch $T_n$, $\sum_{j=1}^n Z_j<-u$. Thus $U(T_n)<0$ and $\psi(u)=1$.

Boundary exception: At equality with $c=0$, the equation $c=\lambda E[X]$, together with $\lambda>0$ and $X\geq0$, implies $E[X]=0$, hence $X=0$ almost surely. Then $U(t)=u$ for all $t$, so ruin does not occur. Thus failure of the net profit condition implies certain ruin for the stated nondegenerate models, but not for the zero-claim, zero-premium boundary case.

Worked Example: Negative Drift and Certain Ruin

A hypothetical insurer has claim frequency $\lambda=0.50$ per year, with each claim equal to $\$3{,}000$, and charges premiums at $c=\$1{,}200$ per year. Initial surplus is $u=\$5{,}000$, and there is no investment income. Determine the drift and the ultimate ruin probability.

Method. The claim amount is constant, so $E[X]=\$3{,}000$. Expected claim outgo per year is $\lambda E[X]=0.50(\$3{,}000)=\$1{,}500$. The drift is $c-\lambda E[X]=\$1{,}200-\$1{,}500=-\$300$ per year.

Since $c<\lambda E[X]$, the strong-law argument applies: surplus at claim epochs divided by the number of claims converges almost surely to $c/\lambda-E[X]=\$1{,}200/0.50-\$3{,}000=\$2{,}400-\$3{,}000=-\$600$ per claim. Therefore claim-epoch surplus eventually becomes negative, regardless of the finite initial surplus. The ultimate ruin probability is $\psi(\$5{,}000)=1$.

As a check on the drift calculation, $E[U(10)]=\$5{,}000-(\$300)(10)=\$2{,}000$. This expected value alone does not show certain ruin by year 10; the conclusion is about eventual ruin, established by the long-run claim-epoch argument. A common mistake is to infer a specific ruin time from negative expected surplus.

Worked Example: Zero Drift with Nondegenerate Claims

A hypothetical insurer has $\lambda=0.25$ claims per year. Claim amounts are exponential with mean $\$4{,}000$, and premium income is $c=\$1{,}000$ per year. Initial surplus is $u=\$2{,}000$, with no investment income. Determine whether the model has positive, zero, or negative drift and whether ultimate ruin is certain.

Method. The exponential claim distribution has $E[X]=\$4{,}000$ and $\operatorname{Var}(X)=(\$4{,}000)^2=\$16{,}000{,}000$. Expected claim outgo per year is $\lambda E[X]=0.25(\$4{,}000)=\$1{,}000$, exactly equal to $c$. The drift is zero.

The claim amounts are nondegenerate, and $c>0$, so the equality case above applies. To check the increments, $E[W_j]=1/\lambda=4$ years and $\operatorname{Var}(W_j)=1/\lambda^2=16$ years squared. Thus $E[Z_j]=cE[W_j]-E[X]=(\$1{,}000)(4)-\$4{,}000=0$, and $\operatorname{Var}(Z_j)=c^2\operatorname{Var}(W_j)+\operatorname{Var}(X)=\$16{,}000{,}000+\$16{,}000{,}000=\$32{,}000{,}000$.

The increment variance is finite and positive. Its partial sums therefore have arbitrarily large negative values almost surely, so they eventually fall below $-\$2{,}000$. Hence the surplus is negative at some claim epoch and $\psi(\$2{,}000)=1$. The common mistake is to argue that zero expected drift means surplus stays at its starting value; the expectation stays constant, but individual paths fluctuate.

Interpreting the Condition Correctly

The net profit condition is a long-run benchmark, not a finite-horizon ruin calculation. A positive drift does not establish a particular value of $\psi(u)$, nor does it eliminate the possibility of ruin. Conversely, when the condition fails in the nondegenerate models considered here, certain ultimate ruin means probability one over an unlimited time horizon; it does not mean that every path is ruined by a specified date.

A reliable exam solution separates three steps: calculate $\lambda E[X]$ in currency per unit time, compare it with $c$ in the same units, and then state what the comparison implies under the model assumptions. In particular, equality is not the same as negative drift, and the zero-claim boundary case must not be silently included in a claim that nonpositive drift always causes ruin.

Compute expected claim outgo.
Multiply claim frequency $\lambda$ by mean claim amount $E[X]$; check that the result has the same time unit and currency unit as $c$.

Compare rates.
If $c>\lambda E[X]$, drift is positive. If $c<\lambda E[X]$, drift is negative. If they are equal, drift is zero.

Apply the correct conclusion.
Negative drift gives certain ultimate ruin by the strong law. Zero drift gives certain ultimate ruin for the nondegenerate model described above; check separately for zero claims and zero premium.

Check Your Understanding

For each hypothetical model, identify the drift and state the conclusion justified about ultimate ruin. Assume no investment income.

A portfolio has $\lambda=0.10$ claims per year, $E[X]=\$6{,}000$, premium rate $c=\$700$ per year, and initial surplus $\$4{,}000$. Does the net profit condition hold? (Answer: Expected claim outgo is $\$600$ per year; drift is $+\$100$ per year, so the net profit condition holds, but this alone does not imply zero ruin probability.)
A hypothetical cover has $\lambda=0.40$, constant claim amount $\$2{,}500$, premium rate $\$900$ per year, and finite initial surplus. What is the drift and ultimate ruin probability? (Answer: Claim outgo is $\$1{,}000$ per year, drift is $-\$100$ per year, and ultimate ruin probability is 1.)
Claim frequency is $\lambda=0.20$ per year, claim amounts are exponential with mean $\$5{,}000$, and $c=\$1{,}000$ per year. What is the drift and ultimate ruin probability for any finite initial surplus? (Answer: Drift is zero; because claims are nondegenerate and $c>0$, ultimate ruin probability is 1.)
Claims occur at rate $\lambda=0.30$ per year but have amount zero almost surely; premium rate and initial surplus are both zero. Does failure of the net profit condition imply ruin here? (Answer: No. Surplus remains zero at all times, so strict-negative surplus ruin never occurs and $\psi(0)=0$.)
A model has $\lambda=0.10$ per year, exponential claim amounts with mean $\$3{,}000$, premium rate $\$250$ per year, and initial surplus $\$1{,}000$. What is the expected surplus after eight years, and what is the ultimate ruin probability? (Answer: Drift is $\$250-\$300=-\$50$ per year; $E[U(8)]=\$1{,}000-8(\$50)=\$600$, and ultimate ruin probability is 1.)

What You'll Learn