Deriving the Adjustment Coefficient Equation

Why the Adjustment Coefficient Equation Has This Form

In the classical risk model, the insurer’s surplus rises continuously through premium income and falls when claims occur. A useful way to study the model is to test an exponential function of surplus. Requiring that this function have no expected drift leads to an equation involving the claim-size moment-generating function. The positive solution, when it exists, is the adjustment coefficient used in exponential upper bounds for ultimate ruin.

The derivation is important because it identifies exactly where claim frequency, premium income, and the claim-size distribution enter. It also helps prevent a common error: omitting the adjustment coefficient from the premium term or using a moment-generating function that is infinite at the proposed root.

Model and notation: Let \(U(t)=u+ct-S(t)\) be the surplus at time \(t\), where \(u\geq 0\) is initial surplus, \(c>0\) is the premium rate per unit time, and \(S(t)=\sum_{j=1}^{N(t)}X_j\) is aggregate claims. Assume \(N(t)\) is a Poisson process with rate \(\lambda>0\), the claim amounts \(X_j\) are independent and identically distributed nonnegative random variables, and the claim amounts are independent of \(N(t)\). Define \(M_X(r)=E[e^{rX}]\) wherever this expectation is finite.

The usual setting for an adjustment coefficient is the net profit condition \(c>\lambda E[X]\), with \(E[X]<\infty\). This says that premium income per unit time exceeds expected claim outgo per unit time. The equation can be written formally without this condition, but a positive root used in ruin bounds is normally sought under the net profit condition. Finiteness of \(M_X(r)\) at the proposed positive value \(r\) is also necessary.

Deriving the Equation from an Exponential Trial Function

Take the trial function \(f(u)=e^{-ru}\) for \(r\geq 0\). This function is positive and decreases as surplus increases. If a claim of size \(X\) occurs, its value at the new surplus \(u-X\) is \(e^{-r(u-X)}=e^{-ru}e^{rX}\). That increase in the trial function is why the claim-size moment-generating function appears with a positive argument.

Apply the surplus drift.
Between claims, surplus increases at rate \(c\), so the drift contribution to the infinitesimal change in \(f\) is \(cf'(u)=-cr e^{-ru}\).

Apply the claim jumps.
Claims arrive at rate \(\lambda\). The expected change in \(f\) at a claim is \(E[f(u-X)-f(u)]=e^{-ru}(M_X(r)-1)\), provided \(M_X(r)\) is finite.

Set the total expected drift to zero.
Adding the drift and jump contributions gives \(e^{-ru}\{-cr+\lambda[M_X(r)-1]\}\). An exponential trial function with zero infinitesimal expected drift must satisfy the adjustment coefficient equation.

\boxed{\lambda\bigl(M_X(r)-1\bigr)=cr}

The same result follows directly from the compound Poisson model. Conditional on \(N(t)=n\), the sum of \(n\) independent claim amounts has exponential moment \(M_X(r)^n\). Averaging over the Poisson count gives

E[e^{rS(t)}] =\sum_{n=0}^{\infty}e^{-\lambda t}\frac{(\lambda t)^n}{n!}M_X(r)^n =\exp\{\lambda t(M_X(r)-1)\}.

Since \(U(t)=u+ct-S(t)\), it follows that

E[e^{-rU(t)}] =e^{-ru}\exp\{t[-cr+\lambda(M_X(r)-1)]\}.

Thus the exponential trial function has constant expectation over time when the exponent multiplying \(t\) is zero, which is exactly the displayed equation. This is a derivation of the equation, not by itself a proof of a ruin bound; establishing such a bound requires an additional argument.

Definition: When it exists, a positive solution \(R>0\) of \(\lambda(M_X(R)-1)=cR\) is called the adjustment coefficient. The value \(r=0\) always satisfies the equation because \(M_X(0)=1\). That trivial root is not the adjustment coefficient used in an exponential ruin bound.

If the claim mean is finite, the function on the left minus the right has derivative at zero \(\lambda E[X]-c\). Under the net profit condition this derivative is negative, so the equation initially moves below zero as \(r\) increases from zero. A positive root, if present, is a nonzero crossing back to zero. Whether such a root exists and is unique is a separate question; the equation alone does not justify assuming either property.

Worked Applications

Worked Example: A Two-Point Claim Distribution

A hypothetical insurer has Poisson claim frequency \(\lambda=1\) per year. Claim amounts, measured in thousands of dollars, are \(1\) or \(2\), each with probability \(1/2\). The premium rate is \(c=2/\ln 2\) thousand dollars per year. Derive the adjustment coefficient equation, check the net profit condition, and verify a positive solution.

Step 1: Write the claim-size moment-generating function. For \(X=1\) or \(2\), each with probability \(1/2\),

M_X(r)=\tfrac12 e^r+\tfrac12 e^{2r}.

Step 2: Form the equation. Substituting \(\lambda=1\) and \(c=2/\ln 2\) into \(\lambda(M_X(r)-1)=cr\) gives

\tfrac12(e^r+e^{2r})-1=\frac{2r}{\ln 2}.

Step 3: Check the net profit condition. The expected claim amount is \(E[X]=\tfrac12(1)+\tfrac12(2)=1.5\) thousand dollars. Expected claims per year are \(1(1.5)=1.5\) thousand dollars. Since \(c=2/\ln 2\approx 2.8854\) thousand dollars per year, \(c>1.5\).

Step 4: Verify the candidate. Let \(R=\ln 2\). Then \(e^R=2\) and \(e^{2R}=4\), so the left side of the equation is \(\tfrac12(2+4)-1=2\). The right side is \(2R/\ln 2=2\). Therefore \(R=\ln 2\approx0.6931\) per thousand dollars is a positive solution. The verification checks the equation; it does not, on its own, prove uniqueness.

Method and trap avoided. The method is to compute \(M_X(r)\), substitute it into the general equation, and then check a candidate root. The exponential uses the claim amount \(X\), not the claim probability or the premium rate.

Worked Example: Continuous Claim Amounts and a Chosen Root

A hypothetical portfolio has \(\lambda=0.5\) claims per year. Each claim amount \(X\), in thousands of dollars, is uniformly distributed on \([0,2]\). The premium rate is \(c=(3-2\ln 2)/(4(\ln 2)^2)\) thousand dollars per year. Derive the equation and verify \(R=\ln 2\).

Step 1: Find the moment-generating function. The density is \(1/2\) on \([0,2]\). For \(r\ne 0\),

M_X(r)=\frac12\int_0^2 e^{rx}\,dx =\frac{e^{2r}-1}{2r}.

At \(r=0\), the limiting value is \(M_X(0)=1\), as required for any moment-generating function.

Step 2: Substitute into the adjustment coefficient equation.

\frac12\left(\frac{e^{2r}-1}{2r}-1\right) =\frac{3-2\ln 2}{4(\ln 2)^2}\,r.

Step 3: Check the net profit condition. For a uniform distribution on \([0,2]\), \(E[X]=(0+2)/2=1\) thousand dollars. Expected claims per year are \(0.5(1)=0.5\) thousand dollars. Writing \(L=\ln 2\approx0.6931472\), the premium rate is

c=\frac{3-2L}{4L^2} =\frac{1.6137056}{1.9218121} \approx0.8397

thousand dollars per year, which exceeds \(0.5\).

Step 4: Verify the candidate. At \(r=L\), \(e^{2L}=4\), so \(M_X(L)=(4-1)/(2L)=3/(2L)\). The left side of the equation is

\frac12\left(\frac{3}{2L}-1\right) =\frac{3-2L}{4L}.

The right side is \(cL=[(3-2L)/(4L^2)]L=(3-2L)/(4L)\). The two sides agree, so \(R=\ln 2\approx0.6931\) per thousand dollars is a positive solution.

Method and trap avoided. Integrating the density gives the moment-generating function, which is then substituted into the equation. Do not substitute only the mean claim amount for \(M_X(r)\); the equation depends on the full claim-size distribution.

Worked Example: Checking Units in a Discrete Portfolio

A hypothetical insurer has \(\lambda=2\) claims per year. Claim amounts in thousands of dollars are \(0.5\) with probability \(0.75\) and \(1.5\) with probability \(0.25\). Its premium rate is \(c=2(1.25\sqrt{2}-1)/\ln 2\) thousand dollars per year. Derive the equation, verify \(R=\ln 2\), and explain a common incorrect calculation.

Step 1: Calculate the moment-generating function and equation.

M_X(r)=0.75e^{0.5r}+0.25e^{1.5r}, \qquad 2\left(0.75e^{0.5r}+0.25e^{1.5r}-1\right)=cr.

Step 2: Check the expected claims and premium. The expected claim amount is \(0.75(0.5)+0.25(1.5)=0.375+0.375=0.75\) thousand dollars. Expected claims per year are \(2(0.75)=1.5\) thousand dollars. With \(L=\ln 2\approx0.6931472\), the premium rate is \(2(1.25\sqrt{2}-1)/L\approx2.2153\) thousand dollars per year, so the net profit condition holds.

Step 3: Verify the positive root. At \(r=L\), \(e^{0.5L}=\sqrt{2}\) and \(e^{1.5L}=2\sqrt{2}\). Therefore

M_X(L)=0.75\sqrt{2}+0.25(2\sqrt{2})=1.25\sqrt{2}.

The left side of the equation is \(2(1.25\sqrt{2}-1)\). The right side is \(cL=2(1.25\sqrt{2}-1)\), so \(R=L\approx0.6931\) per thousand dollars satisfies the equation.

Common mistake and correction. An incorrect calculation sets the left side equal to \(c\), giving \(2(M_X(r)-1)=c\). That drops the factor \(r\). The correct right side is \(cr\): \(c\) is measured in thousands of dollars per year, while \(r\) is measured per thousand dollars, so \(cr\) has units of inverse years, matching the left side.

When the Equation Does Not Give a Finite Positive Root

Worked Example: A Claim Distribution with No Positive Exponential Moment

A hypothetical insurer has \(\lambda=1\) claim per year and premium rate \(c=1.4\) units of money per year. Claim amounts have survival function \(P(X>x)=(1+x)^{-2}\) for \(x\geq0\). Determine whether the adjustment coefficient equation can have a finite positive solution.

Step 1: Check the mean and net profit condition. For a nonnegative random variable, \(E[X]=\int_0^\infty P(X>x)\,dx\). Here

E[X]=\int_0^\infty(1+x)^{-2}\,dx =\left[-(1+x)^{-1}\right]_0^\infty=1.

Thus \(\lambda E[X]=1\), and \(c=1.4>1\); the net profit condition holds.

Step 2: Check the positive exponential moment. The density is \(f_X(x)=2(1+x)^{-3}\) for \(x\geq0\). For every \(r>0\),

M_X(r)=\int_0^\infty \frac{2e^{rx}}{(1+x)^3}\,dx=\infty,

because exponential growth in the numerator eventually dominates the polynomial decay in the denominator, leaving a nonintegrable tail. The left side \(\lambda(M_X(r)-1)\) is therefore infinite for every positive \(r\); it cannot equal the finite quantity \(cr\). There is no finite positive solution to this moment-generating-function equation for these claims, despite the net profit condition.

Method and trap avoided. Check that the moment-generating function is finite at positive arguments before searching for a root. A finite mean alone does not guarantee that the adjustment coefficient equation has a finite positive solution.

Interpreting the Positive Root

The adjustment coefficient is a rate of exponential decay with respect to the monetary scale of surplus. If \(X\) is measured in thousands of dollars, then \(R\) is measured per thousand dollars, so \(Ru\) is dimensionless when \(u\) is also expressed in thousands of dollars. In the usual ultimate-ruin bound, the exponential term has the form \(e^{-Ru}\); a larger \(R\) makes that bound decrease more quickly as initial surplus increases. The equation identifies the candidate coefficient but does not itself establish the bound.

Quantity	Role in the equation	Typical units
\(\lambda\)	Claim arrival rate	Claims per unit time
\(M_X(r)-1\)	Effect of claim sizes at exponential argument \(r\)	Dimensionless
\(c\)	Premium income rate	Money per unit time
\(r\)	Exponential rate applied to claim amount	Inverse money

A reliable exam calculation follows a fixed order: specify the surplus model and money units, compute \(M_X(r)\), write \(\lambda(M_X(r)-1)=cr\), check the net profit condition, and verify any proposed positive root by substitution. The equation always has the zero root; report a nonzero positive solution when one is established, and do not infer existence or uniqueness merely from the form of the equation. This topic is aligned with the SOA Exam ASTAM (Advanced Short-Term Actuarial Mathematics) syllabus.

Check Your Understanding

For each hypothetical model, write or assess the adjustment coefficient equation using the stated units.

A portfolio has \(\lambda=0.8\) claims per year, premium rate \(c=1.6\) thousand dollars per year, and claim amounts in thousands of dollars equal to \(0.5\) with probability \(0.4\) and \(1.5\) with probability \(0.6\). Write the equation and check the net profit condition. (Answer: \(M_X(r)=0.4e^{0.5r}+0.6e^{1.5r}\), so \(0.8[0.4e^{0.5r}+0.6e^{1.5r}-1]=1.6r\); \(\lambda E[X]=0.8[0.4(0.5)+0.6(1.5)]=0.88<1.6\).)
A portfolio has \(\lambda=0.6\) claims per year, certain claim amount \(X=2\) thousand dollars, and \(c=1.5\) thousand dollars per year. Write the equation and determine whether \(r=0.5\) per thousand dollars satisfies it. (Answer: \(0.6(e^{2r}-1)=1.5r\); at \(r=0.5\), the sides are \(0.6(e^1-1)\approx1.03097\) and \(0.75\), so it is not a solution.)
Claim amounts have survival function \(P(X>x)=(1+x/2)^{-3}\) for \(x\geq0\), with \(\lambda=1\) per year and \(c=1.4\) units per year. Check the net profit condition and decide whether a finite positive solution can be obtained from the moment-generating-function equation. (Answer: \(E[X]=\int_0^\infty(1+x/2)^{-3}dx=1\), so \(c>\lambda E[X]=1\); but \(M_X(r)=\infty\) for every \(r>0\), so the equation has no finite positive solution.)
Claims are \(1\) or \(3\) units of money with equal probabilities, \(\lambda=1\) per year, and \(c=4/\ln 2\) units per year. Derive the equation and verify \(R=\ln 2\). (Answer: \(\tfrac12(e^r+e^{3r})-1=(4/\ln 2)r\); at \(R=\ln 2\), the left side is \(\tfrac12(2+8)-1=4\), and the right side is \(4\).)
A portfolio has \(\lambda=12\) claims per year, each claim is exactly \(0.25\) thousand dollars, and \(c=4\) thousand dollars per year. Write the equation and explain why \(12(e^{0.25r}-1)=4\) is not correct. (Answer: \(12(e^{0.25r}-1)=4r\); the proposed equation omits \(r\), and \(cr\), not \(c\), has units of inverse years.)

What You'll Learn