Claim-Size Moment-Generating Functions

Why the Claim-Size MGF Matters

In the classical surplus model, claim amounts are nonnegative random variables. A useful way to summarize their distribution is the moment-generating function (MGF), which weights each possible claim amount by an exponential function. That exponential weighting makes the MGF especially sensitive to large claims: even a severity model with a finite mean may not have a finite MGF at any positive argument.

This distinction matters in ruin theory. Adjustment-coefficient calculations use the claim-size MGF at a positive argument. Before evaluating or solving any such calculation, one must verify that the MGF is finite at the argument in question. Finiteness is a condition for the calculation to be meaningful; it does not, by itself, guarantee that an adjustment coefficient exists.

Definition: For claim amount $X$, its moment-generating function is $$ M_X(t)=E[e^{tX}] $$ at every real $t$ for which this expectation is finite. Its domain is $$ D_X=\{t\in\mathbb{R}:E[e^{tX}]<\infty\}. $$ For nonnegative $X$, $M_X(t)$ is finite for every $t\leq 0$, because $0\leq e^{tX}\leq 1$. The positive part of the domain can depend strongly on the tail of the claim-size distribution.

The argument $t$ has inverse units to $X$. If claim amounts are measured in dollars, $t$ is measured per dollar; if amounts are measured in thousands of dollars, $t$ is measured per thousand dollars. Rescaling claim amounts therefore changes the numerical value of the argument at which the MGF is evaluated.

How to Determine the Domain

For a continuous severity with density $f_X$, calculate the expectation as an integral: $M_X(t)=\int_0^\infty e^{tx}f_X(x)\,dx$, when $X$ is nonnegative. The integral must converge. For a discrete severity, the corresponding sum must converge. The boundary of the domain needs particular care: an MGF may be finite below a boundary and infinite at the boundary itself.

For nonnegative $X$, if $M_X(t)$ is finite at some $t>0$, then it is finite for every $s$ between 0 and $t$, since $e^{sX}\leq e^{tX}$. Thus the positive domain extends from zero up to a boundary, which may or may not be included. Also, $M_X(t)$ is nondecreasing for $t\geq 0$: if $t_2\geq t_1\geq0$, then $e^{t_2X}\geq e^{t_1X}$ for every possible claim amount.

Worked Example: Exponential Claim Amounts

A hypothetical portfolio has individual claim amounts $X$ that are exponentially distributed with rate $\beta=2$ per thousand dollars. Amounts are measured in thousands of dollars, so $f_X(x)=2e^{-2x}$ for $x\geq0$. There is no interest or mortality assumption in this severity model. Find the MGF, its domain, and its value at $t=0.5$.

Method and derivation. Substitute the density into the expectation:

M_X(t)=\int_0^\infty e^{tx}2e^{-2x}\,dx =2\int_0^\infty e^{-(2-t)x}\,dx.

If $t<2$, then $2-t>0$, and the integral converges. Evaluating it gives

M_X(t)=2\left[\frac{-e^{-(2-t)x}}{2-t}\right]_0^\infty =\frac{2}{2-t},\qquad t<2.

If $t=2$, the integrand in the last integral is $2$, so its integral over $[0,\infty)$ is infinite. If $t>2$, the integrand grows exponentially and the integral also diverges. For negative $t$, the expression is finite as well. Therefore $D_X=(-\infty,2)$.

At $t=0.5$, $$ M_X(0.5)=\frac{2}{2-0.5}=\frac{2}{1.5}=\frac{4}{3}=1.3333. $$ The value is rounded to four decimal places. As a check, $M_X(0)=2/2=1$, as every MGF must be at zero. The method avoids the common mistake of treating the boundary $t=2$ as admissible merely because the formula has a denominator that is zero there: the defining integral diverges at that point.

MGF Domains Depend on the Tail

The exponential example has a finite positive domain, but other distributions behave differently. A bounded claim amount has a finite MGF for every real argument, because the exponential is bounded over the range of possible claim amounts. A severity with a sufficiently heavy tail can have a finite mean and still have an infinite MGF at every positive argument.

For familiar parametric models, the domain follows from the same convergence check. A gamma severity with shape $k>0$ and rate $\beta>0$ has $M_X(t)=(\beta/(\beta-t))^k$ for $t<\beta$. A mixture of severity models has an MGF equal to the same weighted mixture of their MGFs, but its positive domain is limited by the component with the smallest positive boundary that has nonzero mixture weight.

Worked Example: A Mixture with Different Tail Rates

Suppose a hypothetical insurer models a claim amount $X$, in thousands of dollars, as follows: with probability $0.7$, $X$ is exponential with rate $4$; with probability $0.3$, $X$ is exponential with rate $1.5$. The component is selected once for each claim. Find the domain of $M_X(t)$ and calculate $M_X(1)$. No interest or mortality assumptions are involved.

Method. Condition on which component generated the claim. The MGF of an exponential distribution with rate $\beta$ is $\beta/(\beta-t)$ for $t<\beta$. Thus

M_X(t)=0.7\frac{4}{4-t}+0.3\frac{1.5}{1.5-t}, \qquad t<1.5.

Both terms must be finite. The rate-4 component is finite for $t<4$, while the rate-1.5 component is finite for $t<1.5$. Since the second component has positive weight, it makes the mixture MGF infinite at $t=1.5$ and beyond. The full domain is therefore $D_X=(-\infty,1.5)$.

At $t=1$, calculate each component before weighting:

\frac{4}{4-1}=\frac{4}{3}, \qquad \frac{1.5}{1.5-1}=\frac{1.5}{0.5}=3.

Therefore $$ M_X(1)=0.7\left(\frac{4}{3}\right)+0.3(3) =\frac{14}{15}+\frac{9}{30} =\frac{11}{6}=1.8333. $$ The result is rounded to four decimal places. As a check, the mixture mean is $0.7(1/4)+0.3(1/1.5)=0.175+0.2=0.375$ thousand dollars, consistent with the component weights summing to one. A common error is to use the faster rate $4$ as the domain boundary because it is the larger rate. The correction is to check every component with positive weight: the slower-decaying component controls the positive boundary.

The MGF also links the distribution to its ordinary moments, but only when the required derivatives are justified. If $M_X(t)$ is finite on an open interval containing zero, then its derivatives at zero give moments: $M_X^{(n)}(0)=E[X^n]$ when the relevant derivative exists. This local condition around zero is stronger than merely knowing a few moments by other means. In particular, having a finite mean does not imply that $M_X(t)$ is finite for any $t>0$.

Worked Example: Finite Mean but No Positive MGF Argument

A hypothetical catastrophic-claim model has a Pareto claim amount $X$, in thousands of dollars, with minimum amount $x_m=1$ and shape parameter $\alpha=2$. Its density is $f_X(x)=2x^{-3}$ for $x\geq1$. Determine whether its mean is finite and find the positive part of its MGF domain. No interest or mortality assumption is involved.

Mean. Direct integration gives $$ E[X]=\int_1^\infty x(2x^{-3})\,dx =2\int_1^\infty x^{-2}\,dx =2\left[-\frac{1}{x}\right]_1^\infty=2. $$ Thus the mean is finite and equals 2 thousand dollars.

Positive MGF arguments. For any $t>0$, $$ M_X(t)=2\int_1^\infty e^{tx}x^{-3}\,dx. $$ To see that this diverges, for $x\geq1$ use $e^{tx}\geq (tx)^4/4!$, which follows from the nonnegative terms in the exponential series. Then $$ 2e^{tx}x^{-3}\geq \frac{2t^4}{4!}x. $$ The integral of the lower bound over $[1,\infty)$ diverges, so $M_X(t)=\infty$ for every $t>0$. For $t\leq0$, $e^{tX}\leq1$, so the MGF is finite. Hence $D_X=(-\infty,0]$.

The check is important: this model has a finite mean but no finite positive MGF argument. The common mistake is to infer positive-domain finiteness from $E[X]<\infty$. The calculation shows why that inference fails: exponential weighting eventually outweighs the polynomially decreasing Pareto density.

Domain Checks in Ruin Calculations

In the classical compound Poisson surplus model, the claim-size MGF is evaluated at a positive argument when carrying out adjustment-coefficient calculations. The domain check is therefore a model check, not a cosmetic detail. A proposed positive argument must lie where $M_X(t)$ is finite. A numerical root finder cannot make an infinite or undefined MGF value valid, and a formula simplified from an MGF expression must not be extended beyond the convergence domain from which it was obtained.

The domain alone does not establish that an adjustment coefficient exists, is unique, or has a particular value. Those questions require additional conditions on the surplus model and its parameters. At this stage, the essential procedure is to identify the severity distribution, determine its MGF domain, and only then use values of the MGF that are within that domain.

State the severity model and units.
Record the distribution, parameters, and whether claim amounts are in dollars or another unit.

Write the expectation defining the MGF.
Use the density integral or probability-weighted sum, rather than relying only on a remembered closed form.

Check convergence, including the boundary.
Determine where the defining integral or sum is finite; a formula’s apparent algebraic form is not a substitute for this check.

Validate any positive argument.
Confirm it lies inside the domain before using the MGF in an adjustment-coefficient calculation.

Common Exam Traps

Ignoring units. An argument stated per dollar cannot be substituted unchanged into an MGF formula derived for claim amounts in thousands of dollars.
Using the fastest-decaying mixture component. Every component with positive weight contributes; the slowest-decaying component can determine the positive boundary.
Including a boundary without checking it. For an exponential rate $\beta$, the MGF is infinite at $t=\beta$, even though its closed form is often written with a denominator $\beta-t$.
Confusing moments with MGF existence. A finite mean, or even several finite moments, does not guarantee a finite MGF at a positive argument.
Treating domain as an existence theorem. A finite MGF at a positive argument does not by itself prove that an adjustment coefficient exists or satisfies any additional model condition.

Key takeaway: The claim-size MGF is $M_X(t)=E[e^{tX}]$, and its domain is part of the result. In ruin calculations that require a positive argument, establish finiteness at that argument before proceeding; tail behavior, mixture components, and measurement units can all change the answer.

Check Your Understanding

For each problem, identify the domain before evaluating any requested MGF value.

A hypothetical claim amount in thousands of dollars is exponential with rate $5$. Find $M_X(0.2)$ and state the domain. (Answer: $M_X(0.2)=5/4=1.25$; $D_X=(-\infty,5)$.)
A severity is gamma with shape $3$ and rate $6$, in thousands of dollars. Find its MGF domain and $M_X(2)$. (Answer: $D_X=(-\infty,6)$; $M_X(2)=(6/(6-2))^3=27/8=3.375$.)
A claim amount is a mixture: with probability $0.4$ it is exponential with rate $2$, and with probability $0.6$ it is exponential with rate $8$. Find the domain and $M_X(1)$. (Answer: $D_X=(-\infty,2)$; $M_X(1)=0.4(2)+0.6(8/7)=52/35\approx1.4857$.)
A Pareto severity has minimum $2$ thousand dollars and shape $3$, with density $3(2^3)x^{-4}$ for $x\geq2$. State whether its mean is finite and give its MGF domain. (Answer: $E[X]=3$ thousand dollars; $D_X=(-\infty,0]$, so there is no finite positive MGF argument.)
A bounded claim amount is $100$ dollars with probability $0.25$ and $300$ dollars with probability $0.75$. Find its mean, state its MGF domain, and give $M_X(0.01)$. (Answer: $E[X]=250$ dollars; $D_X=\mathbb{R}$; $M_X(0.01)=0.25e^1+0.75e^3\approx16.1792$.)

What You'll Learn