Mapping Pricing Assumptions to Surplus Processes

From Pricing Inputs to a Surplus Process

Pricing calculations summarize expected claim costs and premium needs. A surplus process adds a time dimension: it describes how an insurer’s available funds change as premium income arrives and claims are paid. The key modeling task is to translate each input into compatible units before combining them. A claim frequency per policy-year cannot be inserted unchanged into a model whose time variable is measured in months, and a gross premium cannot be treated as fully available to pay claims if expenses are excluded from the surplus definition.

This tutorial builds the standard compound Poisson surplus model from a hypothetical portfolio’s frequency, severity, premium income, and initial surplus. It focuses on representing the assumptions consistently, not on calculating ruin probabilities or deriving conditions for a particular safety loading.

Definition: Let $u$ be initial surplus, $c$ the constant rate of premium income available to the insurer, and $S(t)$ aggregate claims paid by time $t$. The surplus at time $t$ is $$ U(t)=u+ct-S(t). $$ In the compound Poisson model, $S(t)=\sum_{j=1}^{N(t)}X_j$, where $N(t)$ is a Poisson process with rate $\lambda$ per unit of time and the claim amounts $X_1,X_2,\ldots$ are independent and identically distributed, independent of $N(t)$. The empty sum is zero, so $S(0)=0$ and $U(0)=u$.

The model separates three quantities that are easy to conflate. The arrival rate $\lambda$ measures the expected number of claims per unit time. The distribution of $X_j$ describes the amount of one claim. The premium rate $c$ measures dollars of premium income available per unit time. Initial surplus $u$ is a dollar amount at time zero, not a rate.

Translate Exposure and Frequency into Claim Arrivals

Suppose a portfolio has $n$ identical policies in force throughout the period. If each policy has a Poisson claim count with rate $q$ claims per policy-year, and policy claim counts are independent, then the portfolio count has rate $nq$ claims per year. More generally, with exposure $E$ measured in policy-years and a frequency assumption $q$ per policy-year, the expected claim count over that exposure is $Eq$. When exposure accrues at a constant rate, this gives the portfolio arrival rate per unit time.

The result follows from the superposition property of independent Poisson counts. Over a time interval of length $t$, each policy’s count has a Poisson distribution with mean $qt$. The sum of independent Poisson random variables is Poisson, with mean equal to the sum of their means; for $n$ policies, that mean is $nqt$. Thus the portfolio count over length $t$ is Poisson with mean $(nq)t$, which identifies its rate as $nq$.

That conversion depends on the stated exposure and period. A frequency per policy-month must first be multiplied by the number of policies to obtain claims per month, or converted to a yearly rate if $t$ will be measured in years. A frequency observed over a limited period also needs the corresponding exposure; it should not automatically be treated as a full-year rate.

Specify Premium Income and Claim Payments on the Same Basis

For the simple model, take $c$ to be the deterministic premium income that enters surplus per unit time. If the stated premium is gross of expenses but the process is intended to track funds available after expenses, subtract the modeled expense rate before setting $c$. If expenses are omitted, say so and treat the premium as fully available in the model. Do not silently mix a net premium rate with gross claims or a gross premium rate with expenses that have not been accounted for.

The expression $ct$ assumes premium income accrues continuously at a constant rate. Actual policies may pay premiums monthly, quarterly, or at renewal. The continuous-rate expression is an approximation appropriate to the model assumptions. For a question requiring exact installment cash flows, represent the premium payments at their specified times instead of treating them as continuous.

The basic model also assumes no investment income, no interest accumulation on surplus, and no claim inflation or changing exposure unless these are explicitly incorporated. In the examples below, the effective annual interest rate is $i=0\%$; there are no mortality cash flows, because the modeled events are property or service claims rather than life-contingent benefits. Claim amounts are stated in dollars. These assumptions keep the mapping focused on the specified premium and claim cash flows.

Worked Example: Annual Portfolio Frequency and Net Premium

A hypothetical insurer covers 800 devices for a full year. Each device has a Poisson claim frequency of 0.12 claims per device-year. The claim amount $X$ has an exponential distribution with mean $2,500. The annual gross premium is $450 per device, and annual expenses are $90 per device. Initial surplus is $500,000. Map these assumptions to a surplus process with time measured in years, and calculate expected surplus after one year. Assume the portfolio and its rates remain constant. Use $i=0\%$, no mortality cash flows, and no investment income.

Method. Convert the device-level frequency to a portfolio rate, calculate premium income net of the stated expenses, and insert both rates and the initial surplus into $U(t)$. The severity distribution is exponential with mean $2,500 (rate $1/2{,}500$ per dollar); its expected value is therefore $2,500.

The portfolio claim-arrival rate is $$ \lambda=800(0.12)=96\text{ claims per year}. $$

The net premium available per device-year is $$ 450-90=\$360. $$ Therefore the annual premium rate is $$ c=800(360)=\$288{,}000\text{ per year}. $$

The process is $$ U(t)=500{,}000+288{,}000t-\sum_{j=1}^{N(t)}X_j, \qquad N(t)\text{ has rate }96\text{ per year}. $$

Over one year, expected aggregate claims are $96(2{,}500)=\$240{,}000$. Hence expected surplus after one year is $$ E[U(1)]=500{,}000+288{,}000-240{,}000=\$548{,}000. $$

Check. The expected increase in surplus is $48,000, equal to annual net premium less expected annual claims: $288{,}000-240{,}000=48{,}000$. The common mistake avoided here is using gross premiums of $360,000 as $c$ while also treating expenses as reducing available funds.

A Time-Unit Check Before Writing the Process

A surplus equation is only interpretable if each rate uses the unit assigned to $t$. If $t$ is in years, $c$ must be dollars per year and $\lambda$ claims per year. If $t$ is in months, both must be expressed per month. The severity distribution remains a distribution of dollars per claim; changing the time unit does not change the size of an individual claim.

Set the clock.
Choose whether $t$ is measured in years, months, or another stated unit.

Convert exposure and frequency.
Compute the expected portfolio claims per unit of that same time.

Calculate available premium income.
Use the appropriate net or gross premium basis, state how expenses are handled, and express $c$ per unit of the chosen time.

State claim severity and initial surplus.
Specify the distribution of one claim in dollars and the dollar amount $u$ at time zero.

Write and inspect the process.
Use $U(t)=u+ct-\sum_{j=1}^{N(t)}X_j$, then verify that $U(0)=u$ and that all rates match the clock.

Worked Example: Converting Monthly Assumptions to Annual Rates

A hypothetical membership plan has 2,400 active contracts throughout the year. Its monthly claim frequency is 0.0035 claims per contract-month, and individual claim amounts are exponentially distributed with mean $1,800. The gross premium is $14 per contract-month, with expenses of $2.50 per contract-month. Initial surplus is $250,000. Write the process using years as the time unit and find expected surplus after one year. Assume $i=0\%$, no mortality cash flows, and no investment income.

Method. First calculate monthly portfolio rates, then multiply each by 12 to express it per year. The severity distribution is exponential with mean $1,800 (rate $1/1{,}800$ per dollar), so $E[X]=\$1{,}800$.

The monthly portfolio claim rate is $$ 2{,}400(0.0035)=8.4\text{ claims per month}. $$ Thus the annual claim-arrival rate is $$ \lambda=12(8.4)=100.8\text{ claims per year}. $$

Net premium per contract-month is $$ 14-2.50=\$11.50. $$ The monthly portfolio premium income is $2{,}400(11.50)=\$27{,}600$, so the annual rate is $$ c=12(27{,}600)=\$331{,}200\text{ per year}. $$

The annual-time process is $$ U(t)=250{,}000+331{,}200t-\sum_{j=1}^{N(t)}X_j, \qquad N(t)\text{ has rate }100.8\text{ per year}. $$

Expected annual claims are $$ 100.8(1{,}800)=\$181{,}440. $$ Therefore $$ E[U(1)]=250{,}000+331{,}200-181{,}440=\$399{,}760. $$

Check. The expected surplus increase is $149,760, which also equals $331{,}200-181{,}440$. The common mistake avoided is using 8.4 as the annual claim rate even though it is the monthly rate.

Severity, Claim Counts, and the Compound Sum

The aggregate claim amount is random for two reasons: the number of claims is random, and each claim amount is random. The severity variables $X_j$ represent claim payments per loss, not the number of losses. The compound Poisson construction assumes the claim amounts are identically distributed and independent of one another and of the claim-count process. If the portfolio’s claims have materially different severity distributions by segment, a single common severity distribution may not be appropriate without further modeling.

For constant $\lambda$, $N(t)$ has a Poisson distribution with mean $\lambda t$. The aggregate loss by time $t$ is the sum of the first $N(t)$ severities. It is not generally equal to $\lambda t$ times an individual severity: $\lambda t$ is an expected count, not the realized number of claims. Replacing the random sum by an expected amount may be useful for a mean calculation, but it does not specify the surplus path.

For this model, the expected aggregate claims over time $t$ are $\lambda t E[X]$, provided $E[X]$ is finite. Consequently $E[U(t)]=u+ct-\lambda t E[X]$. This expectation is a summary of the process, not a substitute for the process itself: different claim paths can produce different surplus at the same time.

Worked Example: A Mixture Severity Model over Half a Year

A hypothetical insurer continuously covers 5,000 electronic units. Each unit has a Poisson claim frequency of 0.08 claims per unit-year. A claim is $500 with probability 0.70 and $2,500 with probability 0.30, independently from claim to claim. Annual gross premium is $190 per unit and annual expenses are $19 per unit. Initial surplus is $300,000. Write the annual-time process and calculate expected surplus after six months. Assume $i=0\%$, no mortality cash flows, and no investment income.

Method. The severity distribution is the stated two-point mixture. Its mean is $$ E[X]=0.70(500)+0.30(2{,}500)=350+750=\$1{,}100. $$ The frequency and premium assumptions are annual, so no further time conversion is needed.

The portfolio claim-arrival rate is $$ \lambda=5{,}000(0.08)=400\text{ claims per year}. $$

Net premium per unit-year is $$ 190-19=\$171. $$ Thus $$ c=5{,}000(171)=\$855{,}000\text{ per year}. $$

The process is $$ U(t)=300{,}000+855{,}000t-\sum_{j=1}^{N(t)}X_j, \qquad N(t)\text{ has rate }400\text{ per year}. $$

Six months is $t=0.5$ years. Expected claims over that period are $$ 400(0.5)(1{,}100)=\$220{,}000. $$ Premium income over that period is $855{,}000(0.5)=\$427{,}500$. Therefore $$ E[U(0.5)]=300{,}000+427{,}500-220{,}000=\$507{,}500. $$

Check. The expected increase over six months is $207,500, equal to the half-year premium income less expected half-year claims. The common mistake avoided is treating the probability-weighted severity as a claim count; it is the mean dollar amount per claim.

Assumptions and Common Modeling Traps

A compound Poisson surplus process is a compact representation, but it is not automatic that a pricing table supplies all the assumptions it needs. Before interpreting the model, confirm that the claim frequency is appropriate for the exposure, the severity basis matches the claim payments being modeled, and premium income is stated on the same expense basis as surplus. Also confirm that rates are constant over the modeled period and that the assumed independence and Poisson count structure are reasonable for the intended application.

Mixing time units. Multiplying a monthly frequency by annual time without converting the frequency changes the implied expected count. Choose a clock and convert every rate to it.
Using exposure as if it were a rate. The number of policies alone is not an annual claim rate. It must be combined with a frequency per policy-exposure unit.
Confusing gross and net premium. If expenses reduce the funds available to pay claims, account for them before defining $c$, or model expenses separately.
Replacing the random sum with its mean. $\lambda tE[X]$ is expected aggregate claims, not the actual aggregate claims in a realized path.
Assuming the process covers every cash flow. The basic equation omits investment income and any payments not included in $X_j$. Add such cash flows only if the model specifies how they enter.

The process is well specified when the time origin and units are clear, $U(0)=u$, the premium rate is in dollars per time unit, the Poisson rate is in claims per time unit, and the severity distribution describes dollars per claim. Checking these points before further risk analysis prevents a unit error from being mistaken for a substantive conclusion. This modeling discipline is aligned with the SOA Exam ASTAM (Advanced Short-Term Actuarial Mathematics) syllabus.

Check Your Understanding

For each problem, state a process with compatible units. Unless a problem says otherwise, assume $i=0\%$, no mortality cash flows, no investment income, constant exposure, and premium income available to surplus after any stated expenses.

A hypothetical insurer has 1,200 policies, frequency 0.05 claims per policy-year, mean claim severity $4,000, net premium $260 per policy-year, and initial surplus $180,000. Give the annual $\lambda$, $c$, and process, and find expected surplus after one year. (Answer: $\lambda=60$ claims/year, $c=\$312{,}000$/year, $U(t)=180{,}000+312{,}000t-\sum_{j=1}^{N(t)}X_j$, and $E[U(1)]=\$252{,}000$.)
A portfolio has 9,000 policies. Monthly frequency is 0.002 claims per policy-month. Gross premium is $96 per policy-year and expenses are $12 per policy-year. Mean severity is $1,500 and initial surplus is $400,000. Express the process in years and find expected surplus after one year. (Answer: $\lambda=216$ claims/year, $c=\$756{,}000$/year, $U(t)=400{,}000+756{,}000t-\sum_{j=1}^{N(t)}X_j$, and $E[U(1)]=\$832{,}000$.)
A hypothetical portfolio has 1,000 policies, frequency 0.10 claims per policy-year, and net premium $500 per policy-year. Severity is $1,000 with probability 0.80 and $6,000 with probability 0.20. Initial surplus is $75,000. Find the process and expected surplus after three months. (Answer: $E[X]=\$2{,}000$, $\lambda=100$ claims/year, $c=\$500{,}000$/year, and $E[U(0.25)]=\$150{,}000$.)
A fixed portfolio of 600 policies has monthly frequency 0.01 claims per policy-month, net premium $40 per policy-month, mean severity $2,500, and initial surplus $100,000. Write the process using years and find expected surplus after one year. (Answer: $\lambda=72$ claims/year, $c=\$288{,}000$/year, $U(t)=100{,}000+288{,}000t-\sum_{j=1}^{N(t)}X_j$, and $E[U(1)]=\$208{,}000$.)
A hypothetical group has 3,000 policies active for four months. Monthly frequency is 0.004 claims per policy-month; mean severity is $1,000. Net premium is $30 per policy-month, and initial surplus is $50,000. Use months as the time unit and find expected surplus at month four. (Answer: $\lambda=12$ claims/month, $c=\$90{,}000$/month, $U(t)=50{,}000+90{,}000t-\sum_{j=1}^{N(t)}X_j$, and $E[U(4)]=\$362{,}000$.)

What You'll Learn