Health Insurance, Pensions and Employee Benefits Mathematics

Introduction and context

Health insurance, pensions and employee benefits are long-duration financial contracts whose liabilities depend on uncertain events, contingent cash flows, demographic transitions and economic conditions. Unlike a simple life-insurance benefit, a health or employee-benefit obligation may change with morbidity, salary, employment status, medical inflation, provider behaviour and regulation. Actuarial mathematics therefore combines probability, discounting, survival models, stochastic processes, demographic projection and financial economics.

The central actuarial task is to estimate the present value of future benefits and expenses, determine an equitable contribution or premium, establish technical provisions, and assess whether the arrangement remains financially sustainable under adverse experience.

Learning objectives

After studying this chapter, the reader should be able to:

1. Construct expected present values for health, pension and employee-benefit cash flows.
2. Distinguish life-contingent, health-contingent and employment-contingent benefits.
3. Apply multiple-state models to disability, active employment, retirement and death.
4. Calculate pension liabilities and explain pension cost-allocation methods.
5. Apply credibility theory and claims-reserving techniques to health portfolios.
6. Evaluate assumptions, model risk, governance, fairness and regulatory implications.

1. Actuarial valuation framework

Let a benefit cash flow at time t be B_t, with payment-contingency indicator I_t. Under a deterministic effective annual interest rate i, the discount factor is

The expected present value (EPV) at time 0 is

where n is the projection horizon. If benefits are paid continuously, the corresponding expression is

where v(t) is the discount factor for time t. These formulas require consistent payment timing, benefit units and probability definitions. A benefit payable at the beginning of year t uses a different discount exponent from one payable at the end of that year.

For a level annual payment of 1 at the end of each of n years,

is an annuity-due, while

is an annuity-immediate. The distinction is essential in pension valuation.

A premium or contribution may be decomposed as

The expression is a planning identity rather than a universal pricing law. In regulated insurance, capital, profit, taxes, reinsurance and solvency requirements may require additional terms.

2. Health insurance mathematics

2.1 Benefit structures

Health benefits may be:

• indemnity benefits, reimbursing eligible expenses;
• fixed cash benefits, such as a daily hospital allowance;
• managed-care benefits, where payment depends on provider contracts;
• critical-illness benefits, usually paid on diagnosis of a defined condition;
• long-term-care benefits, linked to functional impairment; and
• group benefits, where eligibility depends on employment.

A health claim frequency–severity model represents aggregate claims S as

where N is claim count and X_j is the cost of claim j. If N and severities are independent, with E[N] = λ and E[X] = μ_X,

and

for a Poisson claim count. The variance formula changes if claim counts are overdispersed, severities are dependent, or inflation and utilisation are stochastic.

2.2 Premium rating

A pure risk premium per exposure is approximately

For a portfolio with exposure E, expected aggregate claims are Eλμ_X. A loaded premium can be represented by

where e is a proportional expense rate. If the premium includes a deductible d, the expected payment per claim is E[(X−d)_+], where (y)_+ = max(y,0). With a policy limit u, payment becomes E[min((X−d)_+,u−d)] when u>d.

2.3 Medical inflation and utilisation

If base cost is C_0, medical inflation m and utilisation trend q apply annually, projected cost after t years is

This multiplicative formula assumes constant annual rates and no interaction beyond multiplication. A more realistic model may distinguish price inflation, service intensity, provider mix, demographic ageing and benefit design.

Worked example: health premium

An employer has 2,000 members. Expected annual frequency is 0.18 claims per member and average incurred severity is 4,500 currency units. Expected claims are therefore

2,000 × 0.18 × 4,500 = 1,620,000.

Suppose administration expenses are 8% of premium and a risk margin is 3% of expected claims. If premium P satisfies P = 1,620,000 + 0.08P + 0.03(1,620,000), then

P = 1,668,600 + 0.08P, so P = 1,813,696 approximately.

The average annual premium is approximately 907 currency units per member. This calculation is conditional on the assumed frequency, severity, expenses and risk margin; it is not a substitute for experience monitoring.

3. Health claims reserving

3.1 Why reserves are needed

At valuation date, an insurer owes amounts for reported but unpaid claims (RBNS), incurred but not reported claims (IBNR), claims handling expenses and future payments on open claims. A practical reserve is

The reserve should generally be based on ultimate expected cost, less payments already made, with appropriate allowance for recoveries and discounting where permitted.

3.2 Chain-ladder method

Let C_{i,j} be cumulative claims for origin year i observed at development age j. Age-to-age factors are

The projected ultimate for a cell observed at age k is

and the reserve is ultimate less cumulative paid or incurred to date, depending on the triangle basis. Chain-ladder assumes that historical development patterns are sufficiently representative of future patterns. It is vulnerable to changes in inflation, coding, claims settlement speed, benefit design and mix of business.

Worked reserving example

Suppose cumulative paid claims for three accident years are:

The 12-to-24-month factor is (1,500+1,800) ÷ (1,000+1,200) = 1.5. The 24-to-36-month factor is 1,800 ÷ 1,500 = 1.2. Year 2 ultimate is 1,800 × 1.2 = 2,160, giving a reserve of 360. Year 3 ultimate is 1,400 × 1.5 × 1.2 = 2,520, giving a reserve of 1,120. Total indicated reserve for these two years is 1,480, before considering tail factors, expenses, discounting or judgement.

A Bornhuetter–Ferguson approach may be preferable for immature years when an external expected loss ratio is credible. It combines observed emergence with an a priori ultimate estimate and is less sensitive than pure chain-ladder to random early experience.

4. Multiple-state models for benefits

4.1 States and transitions

A multiple-state model represents an individual’s status as a stochastic process. For a disability income plan, states might be:

1. Active;
2. Disabled and receiving benefit;
3. Recovered and active;
4. Retired; and
5. Dead.

Let μ_{ij}(t) be the transition intensity from state i to state j at time t. If p_i(t) is the probability of occupying state i at time t, the state-probability vector p(t) satisfies

where Q(t) is the intensity matrix, with off-diagonal elements μ_{ij}(t) and diagonal element

For a time-homogeneous model,

The matrix-exponential result applies when intensities are constant over the interval. For time-varying intensities, numerical integration or piecewise-constant approximations are required.

If benefit rate in state i is b_i(t), the continuous expected present value is

This formulation handles benefits payable only while disabled, only while active, or only after retirement. It prevents the common error of applying a single survival probability to a benefit whose eligibility depends on employment or health state.

Figure: Illustrative transitions among employment, disability, retirement and death states.

4.2 Practical implications

Transition intensities should be estimated separately by age, sex where legally and ethically appropriate, occupation, duration in state and benefit definition. Recovery rates often decline with duration, so a constant recovery intensity may be unrealistic. State models also support experience analysis: observed transitions can be compared with expected transitions, and deviations can be translated into reserve or pricing impacts.

5. Pension mathematics

5.1 Defined-contribution and defined-benefit plans

In a defined-contribution plan, the retirement benefit is determined by accumulated contributions and investment returns. If contribution C_t is paid at time t and investment accumulation factor from t to retirement r is A_{t,r}, the retirement account is

In a defined-benefit plan, the sponsor promises a benefit, commonly related to final salary and service. A simplified annual pension is

where α is the accrual rate. Real plans may include career-average earnings, indexation, offsets, spouse pensions and commutation options.

5.2 Pension annuity timing

Suppose a pension of B is paid annually at the end of each year, beginning one year after retirement at age r, continuing while the member is alive. If survival probability from retirement age r to age r+k is {}_k p_r, the life annuity-immediate APV is

This correctly excludes a payment at the retirement date. If the first payment is immediate at retirement, the annuity-due form is instead

The two values differ by the immediate payment B, subject to the precise payment and survival convention. A general pension liability at entry age x, with retirement at r and benefit payable only after retirement, is

for a single retirement benefit conditional on survival to r. More generally, if death, withdrawal, disability and salary progression affect eligibility, the survival term must be replaced by the appropriate state probability and benefit projection.

Worked pension example

A member retires at age 65 and receives 20,000 annually at each year-end while alive. Assume i = 4% and, for illustration, survival probabilities at ages 66, 67 and 68 are 0.99, 0.97 and 0.94; assume no payments after age 68. Then

APV = 20,000(0.99v + 0.97v² + 0.94v³), with v=1/1.04.

Since v ≈ 0.96154, v² ≈ 0.92456 and v³ ≈ 0.88900,

APV ≈ 20,000(0.95192+0.89782+0.83566) = 53,708.

The first payment is one year after retirement. Including an immediate payment would value a different pension design.

6. Pension cost allocation and funding

The projected unit credit (PUC) method allocates each year’s increment in projected final benefit to service periods. At valuation date, the defined-benefit obligation is the present value of benefits attributed to service up to that date, using projected salary where required by the plan.

Under the entry-age normal cost method, the present value of projected benefits is allocated as a level percentage of salary over the period from entry to retirement. The normal cost is the contribution required for the current service year under that allocation. The actuarial liability is the present value of allocated benefits for past service.

A simplified funding balance is

This identity is incomplete for accounting purposes if different measurement bases, smoothing rules, risk adjustments or asset ceilings apply. Funding methods differ in intergenerational equity, contribution stability and sensitivity to salary and interest assumptions.

Important assumptions include mortality, retirement, withdrawal, salary growth, pension indexation, discount rate, expenses and marital or dependant status. Sensitivity analysis should quantify the effect of changes rather than presenting a single deterministic liability as certain.

7. Employee benefits mathematics

Employee benefits include short-term medical benefits, paid leave, bonuses, termination benefits, post-employment medical cover and pensions. Valuation requires separating:

• benefits earned through service already rendered;
• benefits conditional on future service;
• benefits conditional on survival, health or employment;
• vested and non-vested rights; and
• insured obligations and self-insured obligations.

For a benefit payable on termination, the EPV may be written

If the benefit is payable only for involuntary termination, the probability must be specific to that cause. Aggregating voluntary and involuntary termination rates can materially misvalue a benefit when the amounts differ.

For post-retirement medical benefits, expected cost may be modelled as

Medical cost must reflect trend, plan design, cost sharing and the possibility of adverse selection.

8. Credibility theory for health and benefit portfolios

Credibility combines an individual group’s experience with a broader collective estimate. A linear credibility estimate is

where X̄ is observed group experience, m is the collective mean, and 0≤Z≤1. Under Bühlmann credibility,

where n is the number of observation periods, EPV is expected process variance and VHM is variance of hypothetical means. The formula assumes a suitable exchangeable structure and stable parameters; it should not be applied mechanically when a group has changed benefits, underwriting or exposure composition.

Credibility is particularly useful for employer groups with limited claims experience. It reduces volatility while retaining responsiveness to genuine risk differences. Governance should document data quality, homogeneity, exposure weighting and the treatment of catastrophic claims.

9. Integrated illustrative case

Background

An employer with 5,000 employees provides medical insurance, a disability income plan and a defined-benefit pension. Medical claims have recently increased, disability duration has lengthened and the pension discount rate has fallen.

Problem

Management wants to know whether the contribution rate is adequate and whether the reported liabilities reflect emerging risk.

Analysis

The actuary first reconciles exposure, claims and membership data. A health triangle is developed and both chain-ladder and Bornhuetter–Ferguson reserves are produced. A frequency–severity model is stress-tested for medical inflation and high-cost claims. A multiple-state disability model estimates active-to-disabled incidence, recovery and mortality. The pension liability is projected under PUC, with year-end annuity-immediate payments after retirement. Alternative discount rates, mortality improvements, retirement ages and salary growth are tested.

Credibility weighting is applied to the employer’s medical experience because only three years of data are available. The disability model uses duration-dependent recovery rates rather than one constant rate. Pension assets are valued consistently with the selected funding basis.

Approach and outcome

The analysis identifies three separate drivers: medical inflation increases claims cost, slower disability recovery increases duration liability, and a lower discount rate increases pension obligations. Management adopts a contribution increase, introduces case-management controls subject to clinical governance, reviews disability rehabilitation, and establishes quarterly assumption monitoring. The pension committee receives a risk dashboard showing liability duration, funding ratio and stress losses.

Learning points

1. Different liabilities require different probability models.
2. Claims reserves should distinguish development risk from pricing risk.
3. State-transition models avoid inappropriate use of single-state survival probabilities.
4. Pension payment timing must match the contractual design.
5. Governance and data reconciliation are as important as mathematical sophistication.

10. Limitations, governance and contemporary practice

Models are limited by incomplete data, parameter uncertainty, structural change and behavioural response. Medical providers may change coding or treatment intensity; members may alter retirement or claim behaviour; and regulation may redefine benefits. Model risk should be managed through independent review, documentation, validation, version control, back-testing and stress testing.

Data and AI can assist anomaly detection, claims segmentation, reserve diagnostics and mortality or morbidity forecasting. However, predictive accuracy does not remove requirements for explainability, privacy, cybersecurity, fairness and human oversight. Health data are especially sensitive. Proxy discrimination may arise even when prohibited variables are excluded. Actuaries should apply purpose limitation, data minimisation, access controls and documented fairness tests.

Regulatory requirements commonly address solvency, technical provisions, financial reporting, pension funding, consumer protection and data protection. The applicable legal framework depends on jurisdiction; actuarial conclusions must therefore state the governing basis, valuation date and material assumptions.

Best practice includes:

• reconciling data to financial and membership records;
• matching benefit payment timing to policy wording;
• separating incidence, duration, severity and trend assumptions;
• using multiple methods for immature claims;
• performing deterministic and stochastic sensitivity analysis;
• documenting expert judgement and limitations; and
• communicating uncertainty to trustees, management and members.

Chapter Summary

Health insurance and employee-benefit mathematics values contingent cash flows affected by morbidity, mortality, employment, retirement and economic conditions. Aggregate health claims can be analysed through frequency–severity models, while premiums require explicit treatment of expenses, trend and risk margins. Claims reserves include RBNS, IBNR and handling expenses; chain-ladder and Bornhuetter–Ferguson methods have different data requirements and sensitivities.

Multiple-state models represent transitions among active, disabled, retired and dead states and provide a coherent basis for benefit valuation. Pension liabilities depend critically on payment timing: a pension beginning one year after retirement uses an annuity-immediate, not an annuity-due. Defined-benefit funding methods such as projected unit credit and entry-age normal cost allocate liabilities across service periods in different ways. Credibility improves stability for limited experience, while governance, sensitivity analysis, fairness and data protection are essential to responsible actuarial practice.

Key Terms

Actuarial liability: Present value of benefits attributed to service under a specified actuarial method.

Annuity-due: Payments made at the beginning of successive periods.

Annuity-immediate: Payments made at the end of successive periods.

Bornhuetter–Ferguson: Reserving method combining an a priori ultimate estimate with observed emergence.

Chain-ladder: Reserving method extrapolating historical development factors.

Credibility: Method combining individual experience with collective information.

Defined-benefit plan: Plan promising a benefit determined by a formula.

Defined-contribution plan: Plan in which contributions determine accumulated assets and eventual benefit.

EPV: Expected present value of a contingent cash flow.

IBNR: Incurred but not reported claims.

Medical inflation: Increase in the cost of medical services or treatment.

Multiple-state model: Stochastic model describing transitions among health or employment states.

Normal cost: Portion of the actuarial liability allocated to the current service period.

Projected unit credit: Pension method allocating projected benefit to service periods.

RBNS: Reported but not settled claims.

Risk margin: Allowance for uncertainty or required compensation for risk.

State-transition intensity: Instantaneous rate of movement from one state to another.

Technical provision: Amount held to meet expected insurance obligations under a specified regulatory basis.

Review Questions

Short Answer Questions

1. Distinguish an annuity-immediate from an annuity-due.
2. What are RBNS and IBNR reserves?
3. State the assumptions behind the Poisson frequency–severity variance formula.
4. Why can a single mortality probability be inappropriate for disability benefits?
5. Define credibility weighting and explain the role of Z.

Long Answer Questions

1. Explain the actuarial valuation of a post-retirement medical benefit, including trend, survival and eligibility assumptions.
2. Compare chain-ladder and Bornhuetter–Ferguson reserving methods and discuss their limitations.
3. Derive the state-probability equation for a time-homogeneous multiple-state model.
4. Compare projected unit credit and entry-age normal cost methods.
5. Discuss governance, fairness and privacy issues in AI-supported health insurance pricing.

Analytical / Application Questions

1. A pension begins with the first payment one year after retirement. Explain why an annuity-immediate is required and construct the APV expression for a five-year maximum term with survival probabilities.
2. An employer has three years of volatile health claims experience. Design a credibility-based pricing approach and identify data and governance checks.
3. Construct a reserving recommendation when chain-ladder produces a materially higher estimate than Bornhuetter–Ferguson for an immature accident year.

Multiple Choice Questions with Answers

Question 1

A pension payment begins one year after retirement and is paid annually at year-end. Which annuity basis is appropriate?

A. Annuity-immediate

B. Annuity-due

C. Perpetuity-due

D. Accumulation annuity

Correct Answer: A. Annuity-immediate

Question 2

Under a Poisson frequency–severity model with independent claim count N and severity X, which expression equals Var(S)?

A. λE[X]

B. λVar(X)

C. λE[X²]

D. E[N] + E[X]

Correct Answer: C. λE[X²]

Question 3

Using the example’s chain-ladder factors, the Year 2 ultimate claim estimate is:

A. 1,800

B. 2,160

C. 2,520

D. 360

Correct Answer: B. 2,160

Question 4

In a multiple-state model, the diagonal element q_ii of the intensity matrix is:

A. The probability of remaining forever in state i

B. The sum of all incoming intensities

C. The negative sum of outgoing intensities

D. Always equal to zero

Correct Answer: C. The negative sum of outgoing intensities

Question 5

The expected claims cost for 2,000 members, frequency 0.18 and severity 4,500 is:

A. 810,000

B. 1,620,000

C. 1,800,000

D. 9,000,000

Correct Answer: B. 1,620,000

Question 6

In the credibility estimate μ̂ = ZX̄ + (1−Z)m, increasing Z generally causes the estimate to:

A. Move closer to the group’s observed experience

B. Move closer to zero

C. Ignore observed experience

D. Become equal to the variance

Correct Answer: A. Move closer to the group’s observed experience

Question 7

Which method allocates projected defined benefits to service periods using projected salary and service?

A. Projected unit credit

B. Pure premium method

C. Chain-ladder

D. Frequency–severity convolution

Correct Answer: A. Projected unit credit

Question 8

For the example pension, the approximate APV of 20,000 annual payments using the stated three survival probabilities is:

A. 20,000

B. 40,000

C. 53,708

D. 60,000

Correct Answer: C. 53,708

Question 9

Which reserve category refers to claims incurred but not yet reported to the insurer?

A. RBNS

B. IBNR

C. Normal cost

D. Risk margin

Correct Answer: B. IBNR

Question 10

If medical inflation and utilisation trend are constant and multiplicative, projected cost after t years is:

A. C₀ + t(m+q)

B. C₀(1+m+q)^t

C. C₀(1+m)^t(1+q)^t

D. C₀/(1+m+q)^t

Correct Answer: C. C₀(1+m)^t(1+q)^t