Adjustment Coefficient Calculation Pitfalls

Why Adjustment-Coefficient Calculations Go Wrong

The adjustment coefficient summarizes how the classical surplus model’s ultimate ruin probability decays as initial capital increases. Its equation is compact, but it combines claim frequency, claim severity, premium income, and an exponential moment of claim size. An omitted frequency factor or a mismatch of units can produce a plausible-looking answer for the wrong model. Even a correctly rearranged equation can yield algebraic roots that are not valid because the claim moment-generating function is not defined there.

This tutorial focuses on building and checking that equation. All scenarios below are hypothetical. Unless stated otherwise, claim amounts and premium rates are measured in thousands of dollars, and time is measured in years.

Definition: In the classical compound Poisson surplus model, initial surplus is $u$, premium income arrives continuously at rate $c$, and claims arrive according to a Poisson process with rate $\lambda$. Claim amounts $X_1,X_2,\ldots$ are independent and identically distributed, independent of the claim process, and nonnegative. 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 net profit condition is $c>\lambda E[X]$. When a positive solution exists in the domain of the claim moment-generating function, the adjustment coefficient $R$ is the positive solution of $$ \lambda\bigl(M_X(R)-1\bigr)=cR,\qquad M_X(r)=E[e^{rX}]. $$

The units are part of the model, not a cosmetic choice. If $X$ is measured in thousands of dollars, then $R$ is measured per thousand dollars, so $RX$ is dimensionless. The term $\lambda(M_X(R)-1)$ and the term $cR$ must both have units of inverse time. In particular, $c$ is a premium rate per unit of the same time used for $\lambda$, and in the same monetary unit used for $X$.

The equation does not guarantee that a positive solution exists. The net profit condition is necessary for the usual positive adjustment coefficient, but a finite moment-generating function at positive arguments is also needed. If the moment-generating function is infinite for every $r>0$, no positive adjustment coefficient can be obtained from this equation.

Why the Frequency Factor Is There

The factor $\lambda$ arises because the claim count is random. Over a time interval of length $t$, the aggregate claim amount is a compound Poisson sum. Conditional on a claim count $N(t)=n$, its exponential moment is $M_X(r)^n$. Averaging over the Poisson count gives

E\!\left[e^{r\sum_{j=1}^{N(t)}X_j}\right] =\exp\!\left(\lambda t\bigl(M_X(r)-1\bigr)\right),

while the premium contribution to the exponential transform of the surplus contributes the factor $e^{-rct}$. Thus the exponential rate is $\lambda(M_X(r)-1)-cr$. Setting that rate equal to zero gives the adjustment-coefficient equation. This explains why the equation contains $\lambda$: severity alone describes the size of each claim, while $\lambda$ converts the claim-size transform into an aggregate claim rate.

For a nondegenerate claim distribution with a moment-generating function finite near zero, define $g(r)=\lambda(M_X(r)-1)-cr$ wherever the moment-generating function is finite. Then $g(0)=0$ and $g'(0)=\lambda E[X]-c$. Under the net profit condition, this derivative is negative. A positive root, when it exists, is where the curve returns to zero. Convexity helps explain why a positive root is typically unique in the valid domain, but it does not make roots outside that domain meaningful.

Worked Calculations and Their Checks

Worked Example: Exponential Claims With Consistent Monetary Units

A hypothetical insurer has a Poisson claim frequency of $\lambda=2$ claims per year. Claim amounts are exponential with mean $1$ thousand dollars, and premium income is $c=2.5$ thousand dollars per year. Find the adjustment coefficient.

Method and assumptions. The exponential claim distribution has rate $\beta=1$ per thousand dollars, so $E[X]=1$ thousand dollars and $M_X(r)=1/(1-r)$ for $r<1$. The net profit condition holds because $\lambda E[X]=2$ thousand dollars per year and $c=2.5$ thousand dollars per year.

Substitute the frequency, premium rate, and moment-generating function into the equation:

2\left(\frac{1}{1-r}-1\right)=2.5r.

For a positive root, $r\ne 0$. Since $\frac{1}{1-r}-1=\frac{r}{1-r}$, divide both sides by $r$:

\frac{2}{1-r}=2.5,\qquad 1-r=\frac{2}{2.5}=0.8,\qquad R=0.2.

The result is $R=0.2$ per thousand dollars. Check by substitution: $M_X(0.2)=1/0.8=1.25$, so the left side is $2(1.25-1)=0.5$ per year. The right side is $2.5(0.2)=0.5$ per year. The equality holds, and $0.2<1$, so the root is within the moment-generating function’s domain. The common mistake avoided here is mixing dollars and thousands of dollars: using $c=2500$ with claims measured in thousands would not describe this model.

Worked Example: Restoring an Omitted Claim-Frequency Factor

In a hypothetical portfolio, claims occur at rate $\lambda=2$ per year, and every claim is exactly $1$ thousand dollars. Premium income is $c=2/\ln 2$ thousand dollars per year. Find the adjustment coefficient, and check the tempting equation that omits claim frequency.

Method and assumptions. The severity moment-generating function is $M_X(r)=e^r$, finite for every real $r$. Expected claims per year cost $\lambda E[X]=2$ thousand dollars, while $c=2/\ln 2\approx2.8854$ thousand dollars per year. Thus the net profit condition holds.

The correct equation retains the frequency factor:

2(e^r-1)=\frac{2}{\ln 2}r.

Set $r=\ln 2$. Then $e^r=2$, so the left side is $2(2-1)=2$, and the right side is $(2/\ln 2)\ln 2=2$. Therefore $R=\ln 2$, approximately $0.6931$ per thousand dollars. The direct substitution checks both sides without rounding.

The common mistake is to write $e^r-1=cr$, dropping $\lambda=2$. At $r=\ln 2$, that incorrect equation gives a left side of $1$ and a right side of $2$, so the proposed root fails its own check. The omitted factor changes the equation, rather than simplifying it. In general, use $\lambda(M_X(r)-1)=cr$, not $M_X(r)-1=cr$, unless the model actually has $\lambda=1$ in the chosen time unit.

Worked Example: Rejecting an Algebraic Root Outside the Domain

A hypothetical insurer has $\lambda=1$ claim per year. Each claim has a gamma distribution with shape $2$ and rate $1$ per thousand dollars, and premium income is $c=3$ thousand dollars per year. Determine the valid adjustment coefficient and explain why the other algebraic root must be rejected.

Method and assumptions. For this gamma distribution, $E[X]=2$ thousand dollars and $M_X(r)=(1-r)^{-2}$ only for $r<1$. The net profit condition holds because $3>1(2)$. The adjustment equation is

\frac{1}{(1-r)^2}-1=3r,\qquad r<1.

Multiply by $(1-r)^2$, which is nonzero in the stated domain:

$$ 1-(1-r)^2=3r(1-r)^2. $$

Expand both sides: $2r-r^2=3r-6r^2+3r^3$. The root $r=0$ is the trivial root of the original equation, so divide the resulting equation by $r$ to find nonzero candidates:

2-r=3-6r+3r^2,\qquad 3r^2-5r+1=0.

The quadratic formula gives

r=\frac{5\pm\sqrt{13}}{6}, \qquad r_1\approx0.2324,\quad r_2\approx1.4343.

Only $r_1$ is in the domain $r<1$, so $R=(5-\sqrt{13})/6\approx0.2324$ per thousand dollars. Check the equation using the unrounded root: $3R\approx0.6972$, and $(1-R)^{-2}-1\approx0.6972$. The second candidate is not valid: the gamma moment-generating function is infinite at $r_2>1$. The common mistake is to treat every root of the polynomial obtained by clearing a denominator as a valid moment-generating-function solution. A candidate must also belong to the original function’s domain.

A Reliable Calculation Procedure

Before solving numerically or algebraically, make the model and domain explicit. The following sequence catches the most frequent errors.

Fix the units.
Choose one monetary unit for $X$ and $c$, and one time unit for $\lambda$ and $c$. Convert all inputs before substituting.

Check expected claims against premium income.
Calculate $\lambda E[X]$ in money per time and verify $c>\lambda E[X]$. If this fails, do not proceed as though a positive adjustment coefficient were assured.

Write the claim moment-generating function and its domain.
State $M_X(r)$ and exactly which positive values of $r$ make it finite. Keep that restriction beside the equation.

Substitute the complete equation.
Use $\lambda(M_X(r)-1)=cr$, with the actual claim frequency and premium rate. Do not silently set a frequency to one.

Check every proposed positive root.
Discard roots outside the domain, then substitute any remaining root into the original equation and compare both sides.

Exam Traps to Keep in View

Dropping claim frequency. The severity transform $M_X(r)$ concerns one claim. The aggregate claim transform per unit time uses $\lambda(M_X(r)-1)$.
Using premium per policy instead of premium per time. The equation requires the continuous premium income rate $c$, expressed per the model’s time unit.
Mixing monetary scales. If claims are in thousands of dollars, convert a premium rate stated in dollars before using it with those claims.
Forgetting the trivial root. Since $M_X(0)=1$, $r=0$ always satisfies the equation. The adjustment coefficient is the positive root, not zero.
Ignoring the moment-generating-function domain. A numerical solver or polynomial formula can return a number where $M_X(r)$ is infinite or undefined. Such a number is not a solution to the adjustment-coefficient equation.
Assuming the net profit condition alone settles existence. The condition $c>\lambda E[X]$ checks the mean drift, not whether a finite positive exponential moment exists.

Key takeaway: A valid adjustment coefficient is a positive root of $\lambda(M_X(r)-1)=cr$ that lies inside the claim moment-generating function’s domain. Consistent units, the full claim-frequency factor, the net profit check, and substitution into the original equation are all essential parts of the answer. This treatment is aligned with the SOA Exam ASTAM (Advanced Short-Term Actuarial Mathematics) syllabus.

Check Your Understanding

For each problem, write the model equation before using the stated or calculated root.

A hypothetical insurer has exponential claims with mean $0.5$ thousand dollars, claim frequency $2$ per year, and premium income $1.5$ thousand dollars per year. Find the adjustment coefficient in inverse thousands of dollars. (Answer: $R=2/3\approx0.6667$ per thousand dollars.)
Claims are a fixed $2$ thousand dollars each and arrive at rate $1$ per year. Premium income is $2/\ln 2$ thousand dollars per year. Verify the positive adjustment coefficient. (Answer: $R=(\ln 2)/2\approx0.3466$ per thousand dollars.)
Claims have a gamma distribution with shape $2$ and rate $1$ per thousand dollars; $\lambda=1$ per year and $c=3$ thousand dollars per year. Give the valid positive root and state why the other quadratic root is invalid. (Answer: $R=(5-\sqrt{13})/6\approx0.2324$ per thousand dollars; the other root exceeds the mgf domain $r<1$.)
A hypothetical model has fixed claims of $1$ thousand dollars, $\lambda=1$ per year, and $c=0.8$ thousand dollars per year. Does it have a positive adjustment coefficient under the classical equation? (Answer: No; $c<\lambda E[X]=1$, and $e^r-1>r>0.8r$ for $r>0$, so the equation has no positive root.)
Claims are exponential with mean $1$ thousand dollars, arrive at rate $0.5$ per month, and premium income is $1.5$ thousand dollars per month. Find the adjustment coefficient, ensuring that the frequency and premium rates use the same time unit. (Answer: $R=1-0.5/1.5=2/3$ per thousand dollars.)

What You'll Learn