5.5 Experience rating and bonus-malus systems

Experience Rating and Bonus-Malus Systems

Experience rating and bonus-malus systems adjust insurance premiums based on individual claim history rather than relying solely on broad risk class averages. These methods sit at the heart of actuarial pricing: they improve premium accuracy, reduce adverse selection, and create financial incentives for loss prevention.

This section covers the principles behind experience rating, the mechanics of bonus-malus systems (widely used in auto insurance), and the advanced modeling techniques actuaries use to design and evaluate them.

Experience Rating Principles

Experience rating adjusts premiums based on an insured's own past loss experience. Instead of charging everyone in a risk class the same rate, the insurer uses individual data to move premiums closer to each policyholder's true expected cost.

This approach appears across many lines of business: property and casualty, workers' compensation, and group health insurance all rely on experience rating in some form.

Objectives of Experience Rating

• Loss prevention incentives: Policyholders who know their premiums depend on claims have a financial reason to reduce losses.
• Fairness: Premiums reflect actual risk profiles rather than broad averages, so low-risk insureds aren't subsidizing high-risk ones.
• Adverse selection reduction: High-risk insureds can't hide in the average; their premiums rise to reflect their experience.
• Competitive pricing: Insurers can offer customized rates, attracting and retaining lower-risk business.

Prospective vs. Retrospective Experience Rating

These are two distinct timing approaches:

Prospective experience rating sets future premiums based on past loss experience (claim frequency, loss ratios). The insured knows their premium before the policy period begins. This is more common for small to medium-sized risks because it gives policyholders predictability.

Retrospective experience rating adjusts premiums after the policy period ends, based on actual losses incurred during that period. This approach is used for large, complex risks where the volume of exposure is big enough that one period's losses carry meaningful information. It produces more accurate pricing but introduces premium uncertainty for the insured.

Credibility Theory in Experience Rating

Credibility theory answers a fundamental question: how much weight should you give an individual's own experience versus the broader class average?

• Full credibility applies when the insured's own data is large and stable enough to be a reliable predictor on its own.
• Partial credibility blends the insured's experience with the class average. The credibility-weighted premium takes the form:

$P = Z \cdot \bar{X}_{individual} + (1 - Z) \cdot \bar{X}_{class}$

where $Z$ is the credibility factor ($0 \leq Z \leq 1$).

• Credibility $Z$ increases with the volume of data (more exposure years, more claims) and decreases with higher variance in the insured's loss experience.

Bonus-Malus Systems

A bonus-malus system is a specific form of experience rating used primarily in auto insurance. Policyholders are assigned to discrete classes based on their claim history. Each class carries a premium relativity that either discounts ("bonus") or surcharges ("malus") the base premium.

The term comes from Latin: bonus (good) for discounts earned through claim-free periods, and malus (bad) for surcharges triggered by claims.

Definition and Purpose

The system assigns every policyholder to a numbered class. Each class maps to a premium level. The goals are straightforward:

• Reward claim-free policyholders with lower premiums
• Penalize those who file claims with higher premiums
• Improve fairness and reduce adverse selection in the auto insurance market

Structure of Bonus-Malus Systems

A typical system has between 5 and 20+ classes arranged on a scale:

• Base class: Where new policyholders with no history start. This class typically has a relativity of 1.0 (the average risk level).
• Bonus classes (above the base): Relativities less than 1.0, producing premium discounts. A relativity of 0.70, for example, means the policyholder pays 70% of the base premium.
• Malus classes (below the base): Relativities greater than 1.0, producing surcharges. A relativity of 1.50 means the policyholder pays 150% of the base premium.

The relativities for each class are determined through actuarial analysis of claim frequency and severity data.

Transition Rules Between Classes

Transition rules define how policyholders move through the system each policy period:

1. Claim-free year: The policyholder moves up by one (or more) bonus classes.
2. Year with one or more claims: The policyholder drops down by a specified number of classes (commonly 2 or 3 classes per claim, depending on system design).
3. Caps: Most systems impose a ceiling (highest bonus class) and a floor (lowest malus class).

The asymmetry in many systems (dropping faster than climbing) reflects the actuarial reality that a single claim carries more information about risk than a single claim-free year.

Premium Adjustments Based on Class

The premium a policyholder pays equals the base premium multiplied by their class relativity:

$\text{Premium} = \text{Base Premium} \times r_i$

where $r_i$ is the relativity for class $i$. Relativities are calibrated so that the total premium collected across all classes covers expected losses and expenses for the entire portfolio.

Designing Bonus-Malus Systems

Designing an effective system requires balancing multiple objectives: risk differentiation, simplicity, fairness, and profitability. Actuaries use historical claim data, statistical models, and business constraints to inform each design choice.

Determining Number of Classes

The number of classes controls how finely the system segments risk:

• More classes (e.g., 20+) allow finer differentiation between risk levels but increase complexity. Policyholders may find it harder to understand where they stand.
• Fewer classes (e.g., 5-6) are simpler but may not create enough spread between the best and worst drivers to meaningfully influence behavior.

The right number depends on the size of the insured population, the variance in claim frequency, and regulatory or market norms in the jurisdiction.

Setting Transition Rules

Transition rules are the behavioral engine of the system. They determine how strongly the system rewards or penalizes policyholders.

• Strict rules (e.g., drop 3 classes per claim, rise 1 class per claim-free year) create powerful incentives but can feel punitive after a single at-fault accident.
• Lenient rules (e.g., drop 1 class per claim) are more forgiving but may not motivate loss prevention.
• Rules can be symmetric (same magnitude for claims and claim-free years) or asymmetric (larger drops for claims than gains for claim-free years). Asymmetric rules are more common in practice.

Calculating Premium Relativities

Premium relativities are estimated using actuarial models:

1. Analyze historical claim frequency and severity by bonus-malus class.
2. Use generalized linear models (GLMs) to estimate expected claim costs per class, controlling for other rating factors (age, vehicle type, territory, etc.).
3. Set relativities so that the weighted average premium across all classes equals the portfolio's required premium level.
4. Validate relativities against out-of-sample data and update them regularly as new claim experience emerges.

Balancing Incentives and Fairness

The core tension: strong incentives improve loss prevention but can produce outcomes that feel disproportionate.

• A policyholder with one isolated claim in 15 years shouldn't face the same surcharge as someone with three claims in three years. Good system design accounts for this.
• Premium differentials between the highest and lowest classes should be large enough to motivate behavior but not so large that they create affordability problems.
• Excessive premium volatility (large swings year to year) erodes policyholder trust. Smooth transitions matter.

Evaluating Bonus-Malus Systems

Once a system is in place, actuaries need to measure whether it's actually working. Evaluation focuses on risk differentiation, behavioral impact, and financial performance.

Efficiency and Effectiveness Measures

Efficiency captures how well the system separates risk levels:

• The Gini coefficient quantifies risk differentiation. A higher Gini coefficient means the system does a better job distinguishing high-risk from low-risk policyholders.
• The loss ratio (losses divided by premiums) by class reveals whether relativities are well-calibrated. If malus classes consistently show loss ratios above 1.0, their relativities may be too low.

Effectiveness captures behavioral impact:

• Track changes in claim frequency and severity over time after the system is introduced.
• Compare the pure premium (expected losses per exposure) before and after implementation.

Elasticity of Bonus-Malus Systems

Elasticity measures how responsive policyholders are to the system's incentives. High elasticity means policyholders change their behavior (drive more carefully, file fewer marginal claims) in response to premium adjustments. Low elasticity suggests the incentive structure isn't strong enough to influence behavior.

Elasticity can be estimated using statistical models that relate changes in claim frequency to changes in bonus-malus class and the associated premium differential.

Impact on Policyholder Behavior

Beyond aggregate statistics, actuaries examine behavioral responses:

• Do claim frequencies drop after the system is introduced?
• Is there evidence of claim suppression (policyholders not reporting small claims to protect their bonus class)? This is a known side effect that can distort loss data.
• Behavioral economics concepts like loss aversion (people feel losses more strongly than equivalent gains) help explain why malus penalties tend to be more motivating than bonus rewards.

Surveys and focus groups can supplement quantitative analysis with qualitative insights into how policyholders perceive the system.

Comparison with Flat-Rate Pricing

Comparing a bonus-malus system to flat-rate pricing (everyone pays the same base premium) highlights the value of risk-based pricing:

• Low-risk policyholders pay less under bonus-malus; high-risk policyholders pay more.
• Adverse selection decreases because high-risk insureds can no longer hide in the average.
• Loss prevention improves because premiums respond to behavior.

The tradeoffs: bonus-malus systems introduce premium volatility for individual policyholders and higher administrative costs for the insurer.

Practical Considerations

Data Requirements for Implementation

A bonus-malus system requires individual-level claim history data, including both frequency and severity. Insurers need reliable systems for capturing, storing, and analyzing this data over multiple policy periods.

Data quality matters enormously. Missing or inconsistent claim records can distort class assignments and undermine fairness. Actuaries typically build data validation and cleansing processes before launching the system.

Integration with Existing Rating Factors

Bonus-malus class is just one rating variable alongside age, vehicle type, territory, driving record, and others. Integration requires care:

• Avoid double-counting: if driving record (e.g., traffic violations) is already a rating factor, the bonus-malus system should capture residual risk information from claims, not duplicate what's already priced.
• GLMs help isolate the marginal effect of bonus-malus class from other correlated factors.
• The overall rating plan should be reviewed regularly to keep all factors properly calibrated.

Regulatory Constraints and Compliance

Regulatory requirements vary by jurisdiction and can constrain system design:

• Some regulators limit the maximum surcharge or minimum discount.
• Transition rules may be prescribed or restricted.
• Non-discrimination laws may prohibit using certain factors that correlate with bonus-malus class.

Actuaries work with legal and compliance teams to ensure the system meets all applicable requirements before filing for regulatory approval.

Communication to Policyholders

Policyholders need to understand the system for it to influence their behavior. Effective communication includes:

• A clear explanation of the class structure, transition rules, and premium relativities at policy inception
• Annual renewal notices showing the policyholder's current class, how it changed, and what their premium relativity is
• Transparent information about how future claims or claim-free years will affect their class and premium

If policyholders don't understand the system, the behavioral incentives won't work.

Advanced Topics in Bonus-Malus Systems

Optimal Design Using Markov Chains

A bonus-malus system maps naturally onto a Markov chain: each class is a state, and transition probabilities between states depend on claim experience (typically assumed to follow a Poisson process).

With a Markov chain model, actuaries can:

1. Compute the steady-state distribution of policyholders across classes, which determines the long-run premium income.
2. Evaluate how different transition rules affect convergence speed (how quickly the system reaches equilibrium).
3. Optimize premium relativities so that each class's relativity matches its steady-state expected claim cost.
4. Test system stability under different claim frequency assumptions.

The transition matrix $\mathbf{T}$ has entries $t_{ij}$ representing the probability of moving from class $i$ to class $j$ in one period. The steady-state vector $\boldsymbol{\pi}$ satisfies $\boldsymbol{\pi} = \boldsymbol{\pi} \mathbf{T}$.

Incorporating Claim Severity

Traditional bonus-malus systems use only claim frequency (number of claims). This ignores the fact that a policyholder with one large claim may represent a different risk than one with one small claim.

More sophisticated systems incorporate severity by:

• Using claim amount thresholds to determine transitions (e.g., only claims above a certain amount trigger a class drop)
• Jointly modeling frequency and severity using mixed Poisson models for frequency and separate severity distributions
• Applying copulas to capture the dependence structure between claim frequency and severity

These approaches produce more accurate risk differentiation but add modeling complexity.

Bayesian Analysis for Parameter Estimation

Bayesian methods provide a natural framework for bonus-malus parameter estimation because they formalize the blending of prior information with observed data.

• Prior distributions encode expert judgment or historical knowledge about transition probabilities and claim rates.
• As claim data accumulates, the posterior distribution updates to reflect the observed experience.
• MCMC techniques (Gibbs sampler, Metropolis-Hastings) are used to sample from posterior distributions when closed-form solutions aren't available.

This Bayesian updating is conceptually aligned with credibility theory: both blend individual experience with prior expectations. The connection between Bühlmann credibility and Bayesian estimation is a foundational result in actuarial science.

Generalized Linear Models for Bonus-Malus Pricing

GLMs are the standard actuarial tool for multivariate rating, and they extend naturally to bonus-malus pricing:

• The bonus-malus class enters the GLM as a categorical predictor alongside other rating factors.
• For claim frequency, a Poisson or negative binomial GLM with a log link is typical: $\log(\mu_i) = \beta_0 + \beta_1 x_{age} + \beta_2 x_{vehicle} + \cdots + \gamma_{BM\_class}$
• The estimated coefficients $\gamma$ for each bonus-malus class directly yield the premium relativities (after exponentiation).
• GLMs also allow actuaries to test whether the bonus-malus variable adds statistically significant predictive power beyond other rating factors, using deviance tests or AIC/BIC comparisons.