Insurance Pricing, Underwriting and Reserving

Introduction

Insurance pricing, underwriting, and reserving are linked functions that determine whether an insurer can accept risk, charge an adequate price, and meet future claims obligations. Pricing converts expected risk and operating costs into premiums. Underwriting selects and classifies risks before coverage is granted. Reserving estimates the liabilities arising from claims that have already occurred but may not yet be fully reported or settled.

These activities combine finance, statistics, economics, law, and professional judgment. An insurer may appear profitable when premiums are collected, yet become insolvent if prices are inadequate or reserves are understated. Conversely, excessively conservative pricing or reserving can reduce competitiveness and create unfair outcomes for policyholders.

Learning Objectives

By the end of this chapter, students should be able to:

1. Explain underwriting and risk classification.
2. Distinguish exposure, frequency, severity, loss costs, premiums, and technical provisions.
3. Calculate premiums using loss costs, expenses, contingencies, and profit loadings.
4. Explain experience rating and credibility.
5. Describe loss development and major actuarial reserving methods.
6. Evaluate professional, regulatory, ethical, and technological issues in insurance pricing.

1. The Insurance Pricing Cycle

The pricing cycle begins with an exposure: a person, asset, activity, or liability subject to potential loss. The insurer estimates the probability and financial size of losses, adds expenses and risk margins, and determines a premium. Underwriting then decides whether the applicant fits the insurer's acceptance rules. After policies are issued, claims experience is monitored and prices and reserves are reviewed.

A useful distinction is between risk premium and market premium. The risk premium is the amount needed, in expectation, to fund insured losses. The market premium is the amount actually charged after competition, regulation, distribution costs, customer demand, and strategic objectives are considered.

Frequency and severity

Claim frequency measures how often claims occur, while claim severity measures the size of a claim. If there are N claims and E exposure units, observed frequency is:

If aggregate losses are L, observed average severity is:

The pure premium, also called the loss cost, is:

Under a simple frequency–severity model:

The relationship assumes that frequency and severity are measured on a consistent exposure basis and that the product of their expected values is an adequate representation of expected loss cost. If frequency and severity are dependent, or if policy limits and deductibles distort observed claims, more detailed modeling is required.

2. Underwriting and Risk Classification

Underwriting is the process of evaluating an applicant, determining the terms on which insurance may be offered, and controlling the composition of the insurer's portfolio. It includes information collection, risk assessment, classification, policy terms, pricing referral, and post-issuance review.

Underwriting serves four purposes:

• Selection: accepting risks that fit the insurer's appetite.
• Classification: grouping risks with reasonably similar expected loss characteristics.
• Terms and conditions: applying deductibles, limits, exclusions, warranties, and endorsements.
• Portfolio control: preventing excessive geographic, occupational, catastrophe, or counterparty concentration.

Risk classification

A classification variable should be relevant to expected loss, measurable with reasonable reliability, available before or during coverage, and legally and ethically acceptable. Examples include vehicle use, construction type, location, prior claims, age of equipment, and business activity.

A classification system attempts to achieve homogeneity within groups and meaningful differentiation between groups. Perfect homogeneity is impossible. Excessively broad classes create cross-subsidies and adverse selection; excessively narrow classes create complexity, volatility, and possible unfair discrimination.

Important underwriting concepts include:

• Adverse selection: higher-risk applicants are more likely to seek or retain insurance when prices do not reflect their risk.
• Moral hazard: insurance changes behavior in a way that increases the probability or cost of loss.
• Anti-selection: a related process in which policyholders with unfavorable information are more likely to purchase broader coverage or higher limits.
• Risk appetite: the types and amounts of risk an insurer is prepared to accept.
• Referral authority: the threshold above which an underwriter must obtain specialist or management approval.

Underwriting should not be confused with claims assessment. Underwriting evaluates risk before or during coverage; claims assessment determines whether a reported event is covered and the amount payable under the contract.

3. Premium Construction

A premium must normally cover expected claims, operating expenses, the cost of capital and uncertainty, and a return consistent with the insurer's objectives. A general additive expression is:

This form is convenient when components are expressed per policy or exposure unit. A common percentage-loading form is:

where v is the variable-expense ratio, c is the contingency or risk-load ratio, and p is the profit-load ratio. The denominator must be positive. This formula assumes that each percentage is calculated as a proportion of gross premium, not as a proportion of loss cost.

If fixed expense per policy is F and the loss cost is per exposure unit, with E exposure units, then:

This assumes fixed expenses are added after the percentage-loaded amount. Alternative expense conventions produce different formulas, so the basis of every loading must be stated.

Worked example: premium calculation

An insurer estimates an annual loss cost of £420 per policy. Variable expenses are 18% of gross premium, the contingency provision is 4%, the profit provision is 6%, and fixed expense is £35 per policy.

The total percentage loading is 18% + 4% + 6% = 28%. Therefore:

Adding fixed expense:

The premium is approximately £618.33. Its components are not £420 plus 28% of £420, because the stated loadings are percentages of gross premium. If they were explicitly percentages of loss cost, the calculation would instead be £420 × 1.28 + £35 = £573. را

A premium indication may also include trend and exposure adjustments:

The factors must be defined consistently. Trend may reflect inflation, medical-cost changes, repair costs, legal developments, or changes in claim frequency. Historical data must also be adjusted for changes in deductibles, limits, coverage, and case reserving practices.

4. Rating Methods and Credibility

Manual rating and experience rating

Manual rating uses rates developed for a broad classification, often supplemented by deductible, limit, territory, or coverage relativities. It is useful where individual experience is sparse or unreliable.

Experience rating adjusts the premium using the insured's own historical claims. It is more responsive to individual experience but may be unstable when the number of claims is small. A basic experience modification can be represented as:

More advanced methods compare actual losses with expected losses while applying credibility and limits on individual claims.

Credibility

Credibility is the degree of reliance placed on an insured's own experience relative to broader data. A simple credibility-weighted estimate is:

where Z lies between 0 and 1. A higher Z is appropriate when the individual has more credible, stable, and relevant data.

One elementary limited-fluctuation approach sets credibility approximately proportional to the square root of exposure:

where E is the current exposure and E₀ is the exposure required for full credibility. This is an illustrative model, not a universal actuarial standard. It assumes comparable exposure units, a stable underlying risk process, and a selected standard for acceptable statistical fluctuation.

Worked example: credibility

Suppose an employer's own indicated loss cost is £900, the class estimate is £700, and its credibility is 40%. Then:

The result recognizes the employer's experience without allowing a small data set to dominate the estimate.

5. Loss Development and Claims Data

Claims information changes over time. A claim may be incurred but not reported, reported but not fully valued, or paid gradually. Consequently, the amount paid to date is usually less than the ultimate cost.

Key terms are:

• Reported claim: a claim notified to the insurer.
• Paid loss: amount already paid.
• Case reserve: an adjuster's current estimate for a reported but unsettled claim.
• Incurred loss: paid loss plus case reserve.
• IBNR: incurred but not reported claims; in practice, the term may also include development on reported claims not captured in case reserves.
• Ultimate loss: the final estimated cost of claims from a defined accident or underwriting period.

A development triangle organizes claims by accident period and development age. For example, each row may represent an accident year and each column the cumulative paid or incurred amount at 12, 24, and 36 months.

Chain-ladder method

The chain-ladder method estimates development factors from historical age-to-age movements. A volume-weighted factor from development age j to j+1 may be calculated as:

where Cᵢ,ⱼ is cumulative loss for accident period i observed at development age j. The sums include only rows with observations at both ages.

The projected ultimate for a latest observed cumulative amount is:

The formula assumes that historical development patterns are relevant to future claims and that changes in claims handling, inflation, settlement speed, policy mix, and legal conditions have been addressed.

Bornhuetter–Ferguson method

The Bornhuetter–Ferguson method combines an initial expected loss ratio with the proportion of losses expected to emerge. In simplified form:

It is less sensitive than chain-ladder to unusually high or low early reported losses. It is therefore useful for immature periods, new business, or volatile portfolios, but it depends strongly on the quality of the prior expected loss ratio.

6. Technical Provisions and Actuarial Reserving

A technical provision is the amount an insurer recognizes for obligations arising from insurance contracts. Terminology and measurement rules vary by jurisdiction and accounting framework, but the underlying economic idea is consistent: the insurer must recognize the present obligation associated with future claim payments and related expenses.

Common components include:

• Premium provision: expected future claims and expenses arising from the unexpired portion of existing contracts.
• Claims provision: obligations for insured events that have already occurred.
• Outstanding reported claims: estimated unpaid amounts on known claims.
• IBNR provision: estimated unpaid obligations not fully captured in reported claims.
• Risk adjustment or risk margin: compensation for uncertainty in non-financial obligations, where required by the applicable framework.

A simplified claims reserve identity is:

If ultimate claims are estimated at £12 million and £7.5 million has been paid, the claims reserve is £4.5 million. This is not necessarily the same as the accounting liability because discounting, expenses, recoveries, reinsurance, and risk adjustments may be treated separately.

Discounting

When future cash flows are material and the applicable framework permits or requires discounting, the present value of a payment CF_t due at time t is:

where r is the effective discount rate per period. For multiple payments:

The rate, timing convention, inflation assumptions, and treatment of risk must be consistent. Short-tail claims may have immaterial discounting, whereas long-tail liability claims can be highly sensitive to rates and settlement timing.

Best estimate and prudence

A best estimate is generally intended to be an unbiased estimate of expected future cash flows, based on available information and appropriate assumptions. Prudence does not mean arbitrary overstatement. A prudent margin should be transparent, justified, and consistent with the applicable regulatory or reporting basis.

Reserve risk arises from parameter uncertainty, process uncertainty, model uncertainty, and changes in external conditions. Actuaries therefore use multiple methods, sensitivity testing, scenario analysis, and professional judgment rather than relying mechanically on one result.

7. Applied Case: Regional Motor Portfolio

This is an illustrative case.

Background

Harbor Mutual insures 20,000 private motor vehicles. During the previous year, the portfolio produced 1,600 claims and £2.4 million of incurred losses. Average exposure was 20,000 vehicle-years. The insurer expects repair-cost inflation of 5% and forecasts a 3% increase in claim frequency because of urban traffic growth.

Problem

Management must estimate the next year's loss cost and calculate an indicative premium. Variable expenses are 20% of premium, fixed expense is £45 per policy, the contingency provision is 3%, and the profit provision is 5%.

Analysis

Historical frequency is:

Historical severity is:

The projected frequency is £? No: frequency is dimensionless per vehicle-year:

Projected severity is:

Projected loss cost is:

The total percentage loading is 20% + 3% + 5% = 28%. The premium before fixed expense is:

Adding fixed expense gives an indicated premium of approximately:

Approach and governance

Before implementation, Harbor Mutual should test whether the historical losses include catastrophe events, large claims, changes in deductibles, salvage recoveries, and claims-handling changes. It should also examine territory and vehicle-type relativities rather than applying one rate to all policyholders. A governance committee should review fairness, regulatory filing requirements, data quality, and the effect of automated pricing variables.

Outcome and learning points

The calculation produces a transparent starting indication, not an unquestionable final price. Management may adjust it for competition and strategic objectives, but any reduction should be identified as a commercial decision rather than hidden in the actuarial estimate. The case demonstrates the importance of separating frequency, severity, trend, expense basis, and fixed costs.

8. Professional, Ethical, and Regulatory Considerations

Pricing and underwriting are constrained by insurance law, solvency regulation, consumer-protection rules, privacy law, and anti-discrimination principles. Requirements differ across jurisdictions, but common expectations include adequate documentation, actuarial justification, fair treatment, data governance, and the ability to explain material decisions.

Potential ethical problems include:

• using proxy variables that reproduce prohibited discrimination;
• charging vulnerable customers more without a defensible risk rationale;
• withholding material assumptions from regulators or consumers;
• manipulating claims or reserve assumptions to meet earnings targets;
• using personal data beyond the purpose for which it was collected.

Internal controls should include independent review, version-controlled models, data validation, exception reporting, outcome monitoring, and clear accountability. The board and senior management remain responsible for risk appetite and solvency; actuarial or underwriting specialists provide analysis but do not eliminate managerial responsibility.

9. Technology, AI, and Contemporary Developments

Insurers increasingly use telematics, geospatial data, automated document extraction, machine learning, and real-time claims analytics. These tools can improve segmentation and detect fraud, but predictive accuracy alone does not establish fairness or legality.

A responsible model-management process should include:

1. documented data provenance and permitted use;
2. testing for missing data, drift, bias, and instability;
3. out-of-sample validation and stress testing;
4. human review for adverse or exceptional decisions;
5. explainability appropriate to the decision and affected customer;
6. cybersecurity and access controls;
7. periodic review after deployment.

Machine-learning predictions should be reconciled with actuarial reasonableness and business controls. A model that performs well historically may fail after a change in road behavior, climate, medical costs, legislation, or claims operations.

Challenges and Limitations

Insurance estimates are uncertain because losses are random, data may be incomplete, and the future may differ from the past. Inflation, climate change, social inflation, emerging technologies, pandemic effects, judicial decisions, and changes in repair or medical practice can invalidate historical patterns. Small portfolios face high volatility, while large portfolios may conceal important subgroups.

No single method is universally superior. Chain-ladder relies heavily on observed development patterns; Bornhuetter–Ferguson relies on prior expected losses; frequency–severity models require appropriate distributions and dependence assumptions; and expert judgment can introduce bias. Sound practice combines methods, investigates differences, and communicates uncertainty rather than presenting one number as certain.

Best-Practice Checklist

• Define exposure and coverage consistently.
• Separate frequency, severity, trend, and development assumptions.
• State whether loadings are percentages of loss cost or gross premium.
• Adjust historical data for changes in limits, deductibles, inflation, and claims handling.
• Use credibility when individual experience is limited.
• Compare multiple reserving methods and perform sensitivity analysis.
• Reconcile actuarial estimates with accounting and regulatory definitions.
• Maintain independent review and model governance.
• Monitor fairness, customer outcomes, and portfolio concentration.
• Document assumptions, data sources, judgments, and approval responsibilities.

Chapter Summary

Insurance pricing estimates the premium required to fund expected losses, expenses, uncertainty, and profit. Underwriting evaluates applicants, applies risk classification, sets policy terms, and controls portfolio composition. Frequency and severity provide a foundation for calculating loss costs, while expense and profit loadings convert loss costs into gross premiums. Credibility blends individual experience with collective data when an insured's own experience is insufficiently reliable.

Claims do not emerge or settle immediately. Loss development methods estimate ultimate claims from incomplete observations, and actuarial reserving converts those estimates into claims provisions and broader technical provisions. Chain-ladder emphasizes historical development patterns, whereas Bornhuetter–Ferguson combines reported information with an a priori expectation. Discounting, risk margins, reinsurance, recoveries, and accounting rules must be considered when moving from an actuarial estimate to a recognized liability.

Professional pricing and reserving require statistical competence, judgment, documentation, governance, ethical awareness, and continuous monitoring. Technology can improve prediction, but it also creates risks involving bias, privacy, explainability, and model failure.

Key Terms

Underwriting: Evaluation and selection of risks for insurance.

Risk classification: Grouping risks with similar expected loss characteristics.

Exposure: A unit measuring the amount of risk subject to potential loss.

Claim frequency: Number of claims relative to exposure.

Claim severity: Average cost per claim.

Loss cost: Expected loss per exposure unit before expenses and profit.

Gross premium: Amount charged to the policyholder.

Expense loading: Amount included to fund insurer operating expenses.

Profit loading: Amount included to provide a target return.

Adverse selection: Disproportionate participation by higher-risk applicants.

Moral hazard: Risk-increasing behavior resulting from insurance protection.

Credibility: Degree of reliance placed on an experience data set.

Loss development: Change in estimated or paid claims as claims mature.

Case reserve: Estimate for an identified unpaid claim.

IBNR: Claims incurred but not fully reported or recognized in case estimates.

Ultimate loss: Final expected cost of claims for a period.

Technical provision: Recognized liability for insurance obligations.

Best estimate: Unbiased expected value of future insurance cash flows.

Risk margin: Allowance for uncertainty in insurance obligations.

Chain-ladder: Reserving method based on historical development factors.

Review Questions

Short Answer Questions

1. What is the difference between claim frequency and claim severity?
2. Why is risk classification important to an insurer?
3. Define credibility in insurance pricing.
4. What does IBNR mean?
5. Distinguish a case reserve from an ultimate loss estimate.

Long Answer Questions

1. Explain the relationship between underwriting, pricing, and portfolio risk management.
2. Describe the components of a gross premium and explain why the loading basis must be specified.
3. Compare manual rating and experience rating, including their advantages and limitations.
4. Explain the chain-ladder and Bornhuetter–Ferguson reserving methods.
5. Discuss ethical and regulatory risks associated with automated insurance pricing.

Analytical / Application Questions

1. An insurer has a loss cost of £600, variable expenses of 15% of gross premium, a contingency provision of 5%, profit loading of 5%, and fixed expense of £40. Calculate the indicated gross premium and state your assumptions.
2. A portfolio has 2,000 claims and 25,000 exposure units, with total losses of £3,500,000. Calculate frequency, severity, and pure premium. Explain how inflation could affect the next year's estimate.
3. An individual loss-cost estimate is £1,100, the collective estimate is £900, and credibility is 30%. Calculate the credibility-weighted estimate and explain why the collective estimate still has substantial influence.

Multiple Choice Questions with Answers

Question 1

Which measure is calculated as claims divided by exposure units?

A. Claim severity

B. Claim frequency

C. Loss ratio

D. Profit margin

Correct Answer: B. Claim frequency

Question 2

A loss cost is best described as:

A. Expected loss per exposure unit before expenses and profit

B. Total premium collected after expenses

C. The insurer's investment return

D. The unpaid amount on one reported claim

Correct Answer: A. Expected loss per exposure unit before expenses and profit

Question 3

If variable expenses, contingency loading, and profit loading are percentages of gross premium, the appropriate basic formula is:

A. Loss cost × (1 + all loadings)

B. Loss cost ÷ (1 − total percentage loadings)

C. Loss cost − total percentage loadings

D. Loss cost ÷ total percentage loadings

Correct Answer: B. Loss cost ÷ (1 − total percentage loadings)

Question 4

What is the principal purpose of credibility weighting?

A. To eliminate all underwriting judgment

B. To give full weight to the smallest data set

C. To combine individual experience with collective information

D. To convert paid claims into premiums automatically

Correct Answer: C. To combine individual experience with collective information

Question 5

IBNR refers primarily to:

A. Investment income not yet recognized

B. Claims incurred but not fully reported or recognized

C. Premiums written but already earned

D. Expenses paid before policy inception

Correct Answer: B. Claims incurred but not fully reported or recognized

Question 6

Under the simple frequency–severity model, pure premium equals:

A. Frequency plus severity

B. Severity divided by frequency

C. Frequency multiplied by severity

D. Frequency minus severity

Correct Answer: C. Frequency multiplied by severity

Question 7

Which reserving method uses historical age-to-age development factors?

A. Chain-ladder

B. Net present value

C. Break-even analysis

D. Simple average pricing

Correct Answer: A. Chain-ladder

Question 8

If ultimate claims are £10 million and cumulative payments are £6 million, the simplified claims reserve is:

A. £4 million

B. £6 million

C. £10 million

D. £16 million

Correct Answer: A. £4 million

Question 9

A major limitation of relying mechanically on historical development patterns is that:

A. Claims never change after reporting

B. Future claims conditions may differ from historical conditions

C. Premiums cannot be calculated

D. Expenses become zero

Correct Answer: B. Future claims conditions may differ from historical conditions

Question 10

Which practice best supports responsible AI-based underwriting?

A. Using any available personal data without review

B. Ignoring model drift after deployment

C. Combining validation, fairness testing, governance, and human oversight

D. Removing all documentation to protect the model

Correct Answer: C. Combining validation, fairness testing, governance, and human oversight