9. General Insurance Pricing and Claims Reserving

Introduction and Context

General insurance pricing and claims reserving are two sides of the same actuarial task. Pricing determines the premium required for a portfolio of future risks, while reserving estimates the cost of claims arising from policies already written. Both activities depend on incomplete information, uncertain future events, changing economic conditions and behavioural responses to insurance terms.

The central actuarial objective is not simply to calculate an average loss. It is to construct a technically sound estimate of expected cost, allow for expenses and capital requirements, recognise uncertainty, and translate the result into a commercially viable price. A technically adequate premium may nevertheless be unsuitable if it is unaffordable, uncompetitive, vulnerable to adverse selection or inconsistent with the insurer’s risk appetite.

Pricing and reserving therefore require the integration of probability, statistics, financial analysis, claims knowledge, underwriting judgement, data governance and business strategy. The methods examined in this chapter—exposure and loss-cost approaches, rating relativities, experience rating, generalized linear models, chain-ladder, Bornhuetter–Ferguson and stochastic reserving—form a practical toolkit for property, motor, liability, health and other short-tail and long-tail classes.

Learning Objectives

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

1. Explain the relationship between exposure, frequency, severity, pure premium and final insurance premium.

2. Construct rates using exposure-based and loss-cost methods.

3. Interpret rating factors, relativities, experience rating and credibility.

4. Explain the use of generalized linear models in insurance pricing.

5. Prepare and interpret development triangles for claims reserving.

6. Apply the chain-ladder and Bornhuetter–Ferguson methods.

7. Distinguish deterministic from stochastic reserve estimates and assess reserve uncertainty.

8. Allow for claims inflation, changes in exposure, claims handling and social or legal trends.

9. Evaluate profitability, segmentation, governance and ethical risks in pricing and reserving.

1. The Economic Logic of General Insurance Pricing

A general insurance premium must cover more than expected claims. A simplified decomposition is:

Premium = expected claims cost + claims handling expenses + acquisition and administration expenses + cost of capital and profit margin + risk margin.

If expenses are expressed as proportions of premium, the indicated premium can be represented as:

Indicated premium = (expected losses + fixed expenses + risk provision) / (1 − variable expense ratio − commission ratio).

The formula is a useful starting point rather than a complete pricing model. It does not automatically address reinsurance, taxes, catastrophe exposure, investment income, premium payment patterns, policyholder options or competition.

The actuarial distinction between frequency and severity is fundamental. Frequency is the expected number of claims per unit of exposure; severity is the average cost per claim. Expected loss cost is approximately:

Expected loss cost = expected frequency × expected severity.

For example, if a portfolio has 0.08 claims per vehicle-year and average ultimate claim severity of CU 5,000, the expected loss cost is CU 400 per vehicle-year. The final premium will be higher after expenses, uncertainty and profit are considered.

The exposure unit should measure the quantity of risk insured. Examples include vehicle-years, earned car-years, property sums insured, payroll, revenue, passenger-kilometres or policy-years. A poor exposure base can distort both prices and performance comparisons.

2. Exposure, Loss Cost and Rate Indication

2.1 Exposure-based methods

The exposure method estimates the expected cost per exposure unit directly. If ultimate losses are CU 24 million and earned exposure is 60,000 units, the indicated loss cost is CU 400 per unit. This method is especially useful when credible historical loss data are available and the exposure base is relatively stable.

Its principal strengths are transparency and ease of communication. Its weaknesses arise when the portfolio has changed materially, when historical losses are immature, or when the selected exposure measure does not explain risk adequately.

2.2 Loss-cost or pure-premium methods

A loss-cost approach first estimates expected losses per exposure unit and then adds expense provisions, profit and risk margins. In some markets, a loss-cost rate is published or supplied by a rating organisation, and the insurer applies its own expense and underwriting adjustments.

Suppose the projected loss cost is CU 520, fixed expenses are CU 35 per policy, and variable expenses plus commission equal 25% of premium. Ignoring additional risk margins, the indicated premium is:

(520 + 35) / (1 − 0.25) = CU 740.

The distinction between loss cost, technical premium, and market premium is important. The first reflects expected claims, the second incorporates the insurer’s cost structure and risk requirements, and the third is the price actually offered after competitive and strategic considerations.

2.3 Trend and on-level adjustments

Historical losses cannot normally be used without adjustment. A pricing actuary may need to adjust for:

• changes in exposure mix and policy limits;
• inflation in repair, medical and legal costs;
• changes in deductibles and coverage;
• rate changes already implemented;
• changes in claims settlement practices;
• legislative and judicial developments; and
• catastrophe or large-loss experience.

An on-level adjustment restates historical premiums to the current rate level. A trend adjustment projects historical losses to the period in which the new policies are expected to incur claims. Trend should not be applied mechanically: frequency trend, severity trend and exposure trend may differ.

3. Rating Relativities and Portfolio Segmentation

A rating factor expresses how risk varies across categories. If the base expected loss cost is CU 400 and a territory relativity is 1.30, the territorial component is CU 520 before other factors are applied. A multiplicative model might be written as:

Expected loss cost = base rate × territory relativity × vehicle relativity × usage relativity × driver relativity.

Relativities should be estimated from credible data and reviewed for stability. A factor can be statistically significant but commercially or ethically inappropriate. Excessively fine segmentation can create volatile prices, increase operational complexity and produce discriminatory outcomes through proxy variables.

Segmentation has two competing effects. Better segmentation can reduce cross-subsidies and improve risk selection, but it can also make insurance less accessible to higher-risk groups and encourage competitors to target profitable segments. The objective is not maximum segmentation; it is economically justified segmentation subject to legal, social and governance constraints.

Experience rating and credibility

Manual or prospective rating uses broad portfolio information. Experience rating adjusts an individual risk’s premium using its own past claims. A simple credibility formula is:

Credibility-weighted estimate = Z × individual experience + (1 − Z) × collective average.

Here, Z is the credibility factor. It increases when exposure is large, data are reliable and the risk is relatively homogeneous. A small commercial account with one severe claim should not be priced as though that single observation fully represents its future risk. Credibility protects against random fluctuation while retaining information about genuine risk differences.

Experience rating must address large losses, changes in exposure, claim development and the possibility that past claims resulted from unusual circumstances. It can create incentives for prevention, but it may also penalise policyholders for events outside their control or encourage claims reporting behaviour designed to influence renewal terms.

4. Generalized Linear Models in Pricing

Generalized linear models (GLMs) provide a disciplined way to relate claim outcomes to rating variables. A frequency model may use a Poisson or negative binomial distribution with exposure as an offset. A severity model may use a Gamma or lognormal distribution. A compound model combines frequency and severity to estimate pure premium.

For a frequency model:

log(E[N]) = log(exposure) + β₀ + β₁x₁ + β₂x₂ + … + βₖxₖ.

The exposure offset ensures that a policy with twice the exposure is expected to generate approximately twice the claims, all else equal. Exponentiating a coefficient gives an approximate multiplicative relativity. If a coefficient for a category is 0.182, its relativity is exp(0.182), approximately 1.20.

GLMs are valuable because they can control for several factors simultaneously, provide interpretable relativities and support statistical diagnostics. Important diagnostic questions include whether residuals show structure, whether variables interact, whether overdispersion exists, whether exposure is correctly measured and whether extreme observations dominate the estimates.

Model selection should not rely solely on goodness of fit. Out-of-sample validation, stability over time, interpretability, fairness and business plausibility are equally important. A highly predictive variable may be unsuitable if it is legally restricted, acts as a proxy for protected characteristics or cannot be explained to customers and regulators.

Modern insurers may supplement GLMs with gradient boosting, random forests and neural networks. These methods can capture nonlinearities and interactions, but they introduce challenges concerning explainability, data drift, reproducibility and governance. A transparent GLM may remain preferable where pricing decisions must be justified.

5. From Written Premium to Ultimate Claims

Reserving concerns obligations arising from past exposure. The principal reserve categories are:

• case reserves, established on reported claims by claims handlers;
• IBNR reserves, for incurred but not reported claims and development on reported claims;
• unearned premium reserves, for future coverage under policies already written; and
• premium deficiency or liability adequacy reserves, where future costs exceed the value of related unearned premium.

The actuarial reserve generally estimates the ultimate cost of claims less amounts already paid. Ultimate claims include reported payments, future payments on known claims and claims not yet reported.

Claims data are commonly arranged in a development triangle. Rows represent accident years, underwriting years or report years; columns represent development periods; cells contain cumulative paid claims, cumulative incurred claims, claim counts or other measures. The choice of origin period matters. Accident-year triangles are useful for loss occurrence, while underwriting-year triangles align more closely with policy transactions.

6. The Chain-Ladder Method

The chain-ladder method assumes that historical development patterns provide a reasonable guide to future development. For cumulative claims, the age-to-age factor from development period j to j+1 is commonly calculated as:

fⱼ = sum of claims at development j+1 / sum of claims at development j.

The cumulative development factor (CDF) to ultimate is the product of future age-to-age factors. If the latest observed cumulative claim amount is CU 10 million and the selected CDF is 1.25, the projected ultimate is CU 12.5 million and the reserve is CU 2.5 million, subject to any relevant adjustments.

The method is simple, transparent and effective where development patterns are stable. However, it is heavily dependent on the most recent observed claims and can be unreliable for immature accident years, rapidly changing portfolios, sparse data and long-tail classes. Paid and incurred chain-ladder results may differ because incurred data include case reserve judgements, while paid data reflect actual settlement timing.

A sound application requires selection rather than blind mechanical averaging. The actuary should inspect individual development factors, calendar-year diagonals, large claims, operational changes, claim closure rates and shifts in case reserving philosophy. Development factors may need to be adjusted for claims inflation or known changes in settlement speed.

7. Bornhuetter–Ferguson: Combining Prior Expectations with Emerging Experience

The Bornhuetter–Ferguson (BF) method is designed to reduce the volatility of immature observations. It begins with an expected ultimate loss and uses the proportion of losses expected to emerge by the valuation date.

If expected ultimate losses are CU 20 million and 40% is expected to have emerged, the expected reported amount is CU 8 million. If actual reported claims are CU 7 million, the BF ultimate is:

reported claims + expected unreported claims = 7 + (20 × 60%) = CU 19 million.

Equivalently, the unreported portion is based on the prior expected ultimate rather than on the ratio of actual reported claims to development factors. This prevents an unusually low early result from automatically producing an extremely low ultimate estimate, or an unusually high early result from producing an excessive estimate.

BF is especially useful where early development is volatile, data are sparse, a new product has limited experience or a reliable expected loss ratio is available. Its weakness is that the prior expected ultimate can dominate the result, masking genuine deterioration if it is not updated. The choice between chain-ladder, BF and a blend should be justified by data maturity, process changes and the credibility of emerging experience.

8. Stochastic Reserving and Uncertainty

A point estimate is not a complete description of a reserve. Two portfolios can have the same expected reserve but very different uncertainty. Stochastic reserving models generate a distribution of possible outcomes and can provide prediction intervals, percentile reserves and measures such as standard deviation or coefficient of variation.

Approaches include bootstrap chain-ladder, Mack-type models, generalized linear reserving models and simulation models. A bootstrap procedure may resample residuals from a fitted reserving model, simulate process variation and produce many possible ultimate outcomes. The resulting distribution distinguishes, at least conceptually, parameter uncertainty from process uncertainty.

Reserve risk may be summarised by a selected percentile. A 75th-percentile reserve is not automatically a prudent reserve in every context; the appropriate level depends on accounting standards, capital objectives, risk appetite and the purpose of the analysis. Correlation across accident years and lines of business must be considered when aggregating reserve risk.

Uncertainty is often asymmetric. A liability portfolio may have a limited chance of unusually favourable development but a substantial chance of severe adverse development caused by legal inflation, latent claims or large settlements. Management should therefore examine scenarios, stress tests and reverse stress tests, not only model-generated percentiles.

9. Claims Inflation, Social Inflation and Emerging Risk

Claims inflation includes ordinary price inflation, wage inflation, medical inflation, repair-cost inflation and changes in the cost of legal services. Social inflation refers more broadly to rising liability costs associated with litigation financing, changing views of compensation, expanded theories of liability, larger awards and broader plaintiff access to legal remedies.

Inflation affects both pricing and reserving. If a claim payment expected in three years is CU 100,000 today and annual claims inflation is 6%, its nominal projected cost is approximately CU 119,101. Applying a general consumer price index alone may understate specialist medical, construction or legal cost inflation.

Actuaries should distinguish calendar-year effects from accident-year effects, monitor settlement speed and analyse severity by claim type. Scenario analysis is particularly important where historical data do not contain the emerging trend. Climate-related perils, cyber claims, autonomous vehicles, supply-chain disruption and changing regulation may also invalidate historical assumptions.

10. Profitability and Portfolio Management

Pricing performance should be assessed using measures such as the loss ratio, expense ratio and combined ratio:

Combined ratio = loss ratio + expense ratio.

A combined ratio below 100% indicates underwriting profit before consideration of certain capital and investment effects, but interpretation depends on reserving adequacy and the treatment of catastrophe losses. A portfolio may show a satisfactory current ratio while being underpriced if reserves are understated.

More advanced analysis considers return on capital, risk-adjusted profitability, persistency, customer lifetime value, reinsurance cost and concentration risk. Segment-level results should be reviewed for credibility. Small segments can produce extreme ratios through random fluctuation, while large apparently profitable segments may contain hidden catastrophe or liability accumulation.

Pricing, underwriting and claims teams should operate as a feedback system. Pricing assumptions should be tested against actual emerging claims; claims trends should inform underwriting rules; and underwriting changes should be reflected in reserving selections. A cycle of retrospective monitoring, model recalibration and governance review is essential.

Illustrative Case Study: Meridian Motor Insurance

Background

Meridian is a fictional insurer writing private motor insurance. It has 100,000 vehicle-years of exposure. Historical data show 0.075 claims per vehicle-year and average ultimate severity of CU 6,000, giving an indicated expected loss cost of CU 450 per vehicle-year.

Situation

Repair-cost inflation has accelerated, claims settlement has slowed, and the portfolio has shifted toward urban drivers. The existing tariff was based on older regional averages. Meridian’s incurred chain-ladder indicates an ultimate loss ratio of 68%, while a BF analysis using an updated expected loss ratio of 72% indicates a higher ultimate for the two most recent accident years.

Analysis

The chain-ladder result is sensitive to immature incurred development and may be affected by a recent change in case reserving instructions. The BF result gives greater weight to the updated expected loss ratio and is therefore more credible for immature years. A pricing GLM also shows that urban territory, vehicle repairability and annual mileage are material predictors after controlling for driver age.

The actuary recommends separate frequency and severity models, an inflation adjustment based on repair-cost data, and credibility-weighted experience rating for large fleets. A stochastic reserving analysis shows substantial adverse-tail risk from liability claims, even though the central estimate is close to the deterministic selected reserve.

Approach and Outcome

Meridian adopts a blended reserving selection: chain-ladder for mature years, BF for the two newest years and a monitored blend for intermediate years. It updates territorial and vehicle relativities, introduces a claims inflation scenario, and requires governance approval for variables that may operate as socioeconomic proxies. The technical rate increases, but underwriting rules and deductibles are also revised to reduce adverse selection.

After implementation, management monitors actual-to-expected frequency, severity, quote conversion, retention, complaints and segment profitability. The outcome is not judged only by premium growth; it is evaluated through loss emergence, reserve development and risk-adjusted return.

Learning Points

The case illustrates that pricing and reserving cannot be separated from data maturity, claims operations and portfolio change. A single method can be mathematically correct yet inappropriate if its assumptions do not match the business. Blending methods, documenting judgement and testing uncertainty are often more defensible than presenting a mechanically precise answer.

Risks, Governance and Professional Judgement

Common mistakes include using written rather than earned exposure, failing to adjust for rate changes, treating case reserves as unbiased forecasts, ignoring large-loss distortions, extrapolating unstable development factors and confusing statistical significance with commercial relevance. Double counting inflation or applying trends to already trended data is another frequent error.

Governance should define data ownership, model purpose, approval authority, validation standards, override procedures and monitoring thresholds. Material assumptions must be documented, reproducible and subject to independent challenge. Reserve committees should understand not only the selected estimate but also the reasons for changes from the previous valuation.

Ethical practice requires fairness, confidentiality and responsible use of personal data. Automated pricing should comply with applicable insurance, privacy and equality requirements. Policyholders should not be exposed to unjustified or unexplained discrimination. Claims reserving also has ethical significance: deliberate understatement can mislead management, regulators and investors, while excessive conservatism can distort pricing and unfairly transfer costs between policyholders.

Technology and Emerging Practice

Cloud data platforms, automated claims feeds, telematics, geospatial information and document extraction can improve timeliness and granularity. Machine learning may assist fraud detection, claims triage and severity prediction. However, automated systems require controls for data drift, cyber risk, bias, adversarial behaviour and human oversight.

Emerging practice is moving toward integrated pricing and reserving platforms, near-real-time portfolio monitoring, scenario libraries for climate and liability trends, and model ensembles that combine actuarial and machine-learning methods. The actuarial role is consequently shifting from producing a single number toward explaining a range of outcomes, identifying model limitations and supporting accountable decisions.

Best-Practice Framework

A robust pricing and reserving process should:

1. Define the decision, coverage and valuation basis clearly.

2. Reconcile exposure, premium, claims and financial data.

3. Adjust historical information for coverage, mix, inflation and operational changes.

4. Use methods appropriate to data maturity and claim characteristics.

5. Compare alternative methods rather than relying on one output.

6. Quantify uncertainty and perform adverse scenarios.

7. Validate models using out-of-sample and actual-to-expected analysis.

8. Document assumptions, judgement and management overlays.

9. Monitor fairness, legal compliance, data security and model drift.

10. Feed emerging experience back into underwriting, pricing and reserving.

Chapter Summary

General insurance pricing estimates the premium required for uncertain future claims and expenses, while claims reserving estimates the ultimate cost of obligations arising from past exposure. Exposure and loss-cost methods provide the basic rate indication; rating relativities, credibility and experience rating refine it for observed risk differences. GLMs offer an interpretable multivariable framework for frequency, severity and pure-premium modelling.

Reserving requires development of reported and paid claims to ultimate values. Chain-ladder relies primarily on observed development patterns, whereas Bornhuetter–Ferguson combines a prior expected ultimate with the proportion of claims expected to emerge. Stochastic methods extend these approaches by modelling reserve uncertainty and adverse outcomes. Inflation, social trends, portfolio changes and claims operations must be incorporated into both pricing and reserving.

The professional task is broader than calculation. Actuaries and insurance managers must assess profitability, segmentation, fairness, data quality, governance, technology and strategic consequences. Transparent assumptions, alternative methods, monitoring and scenario analysis are central to sound practice.

Key Terms

Exposure: A measurable unit representing the quantity of risk insured.

Frequency: Expected number of claims per unit of exposure.

Severity: Average cost per claim.

Pure premium: Expected claims cost per exposure unit before most expenses and profit provisions.

Loss cost: Expected loss amount allocated to an exposure unit.

Relativity: Multiplicative factor showing how a risk category differs from a base category.

Experience rating: Pricing that incorporates the historical claims experience of the individual risk or account.

Credibility: Weight assigned to an individual risk’s experience relative to collective experience.

Generalized linear model: Statistical model linking an expected outcome to predictors through a link function and specified probability distribution.

Exposure offset: A GLM term that adjusts expected claim counts for differing amounts of exposure.

Case reserve: Claim estimate established for a reported claim.

IBNR: Reserve for claims incurred but not yet reported and further development on reported claims.

Ultimate claim: Total expected cost of a claim or accident period after all development.

Development triangle: Table showing claims or counts by origin period and development period.

Chain-ladder: Reserving method projecting ultimate claims from historical development factors.

Bornhuetter–Ferguson: Reserving method combining actual emerged claims with an expected unreported component.

Cumulative development factor: Product of selected future development factors to ultimate.

Reserve risk: Uncertainty that actual future claim payments will differ from the selected reserve.

Social inflation: Growth in liability claim costs arising from legal, social and litigation-related forces beyond ordinary price inflation.

Combined ratio: Loss ratio plus expense ratio, commonly used to assess underwriting performance.

Review Questions

Short Answer Questions

1. Distinguish claim frequency from claim severity and explain their relationship to expected loss cost.

2. What is the purpose of an exposure offset in a frequency GLM?

3. Define credibility and explain why it is useful in experience rating.

4. What does a cumulative development factor represent?

5. State one important difference between chain-ladder and Bornhuetter–Ferguson reserving.

Long Answer Questions

1. Explain how an insurer moves from historical losses to an indicated premium, including trend, expenses and risk margins.

2. Discuss the construction, interpretation and governance of rating relativities.

3. Describe how a GLM can be used to model claim frequency and severity, including key diagnostic issues.

4. Explain the chain-ladder method and evaluate the circumstances in which its assumptions may fail.

5. Discuss the role of stochastic reserving in management of reserve uncertainty and capital risk.

Analytical / Application Questions

1. A portfolio has expected frequency 0.06 and severity CU 8,000. Fixed expense is CU 40 per policy and variable expenses equal 20% of premium. Calculate the indicated premium before an additional profit margin.

2. Reported claims are CU 7 million, expected ultimate losses are CU 20 million and 40% of claims are expected to have emerged. Calculate the Bornhuetter–Ferguson ultimate estimate and explain why it differs from a simple chain-ladder result.

3. An insurer observes profitable mature accident years but adverse development in the two newest years after a change in claims handling. Recommend a reserving approach and identify the additional investigations required before selecting the reserve.

22. Multiple Choice Questions with Answers

Question 1

A portfolio has an expected frequency of 0.08 claims per vehicle-year and an average severity of CU 5,000. What is the expected loss cost per vehicle-year?

A. CU 40

B. CU 400

C. CU 5,080

D. CU 62,500

Correct Answer: B. CU 400

Question 2

Why is an exposure offset used in a Poisson frequency GLM?

A. To remove all large claims from the data

B. To ensure claim expectations reflect differing amounts of exposure

C. To convert severity into frequency

D. To impose a fixed expense ratio

Correct Answer: B. To ensure claim expectations reflect differing amounts of exposure

Question 3

An individual account has limited claims experience. Which method best prevents one unusual claim from fully determining its renewal premium?

A. Pure individual experience rating

B. Credibility weighting

C. Ignoring all historical information

D. Using only the highest claim

Correct Answer: B. Credibility weighting

Question 4

If a base loss cost is CU 400 and a territory relativity is 1.30, what is the territory-adjusted loss cost before other factors?

A. CU 270

B. CU 400

C. CU 520

D. CU 1,700

Correct Answer: C. CU 520

Question 5

Under the chain-ladder method, a cumulative development factor of 1.25 applied to observed cumulative claims of CU 10 million gives an ultimate estimate of:

A. CU 8 million

B. CU 10.25 million

C. CU 12.5 million

D. CU 25 million

Correct Answer: C. CU 12.5 million

Question 6

In a Bornhuetter–Ferguson calculation, expected ultimate losses are CU 20 million, 40% is expected to have emerged, and reported claims are CU 7 million. What is the BF ultimate estimate?

A. CU 8 million

B. CU 12 million

C. CU 19 million

D. CU 27 million

Correct Answer: C. CU 19 million

Question 7

Which circumstance most strongly supports using Bornhuetter–Ferguson rather than relying exclusively on immature chain-ladder results?

A. Very stable and fully developed claims data

B. Sparse or volatile early development with a credible prior expected loss ratio

C. No available exposure information

D. A portfolio with no expected future claims

Correct Answer: B. Sparse or volatile early development with a credible prior expected loss ratio

Question 8

What does stochastic reserving add to a deterministic reserve estimate?

A. A guarantee of the exact final claims cost

B. A distribution describing possible reserve outcomes and uncertainty

C. A replacement for all actuarial judgement

D. An automatic reduction in claims inflation

Correct Answer: B. A distribution describing possible reserve outcomes and uncertainty

Question 9

Which issue is most directly associated with social inflation?

A. A decline in the number of insured vehicles

B. Higher liability costs caused by legal and litigation-related developments

C. A mechanical reduction in policy limits

D. Lower administrative expenses

Correct Answer: B. Higher liability costs caused by legal and litigation-related developments

Question 10

A combined ratio of 96% generally indicates that, before relevant investment and capital considerations, the insurer has:

A. An underwriting loss of 96%

B. An underwriting profit margin of approximately 4%

C. No claims expense

D. A premium deficiency of 96%

Correct Answer: B. An underwriting profit margin of approximately 4%