Auto Insurance Pricing: Rating Variables and Territory

Key Rating Variables

Personal auto insurance pricing uses numerous rating variables to segment risks and match premiums to expected loss costs. The goal is to charge each policyholder a premium that reflects their individual risk level, ensuring that low-risk drivers are not subsidizing high-risk drivers. Achieving this requires analyzing large datasets to identify which characteristics are predictive of loss frequency and severity.

Driver-related factors include age, gender (in states where permitted), marital status, driving record (accidents and violations), years of licensed driving experience, and education level. Age is one of the strongest predictors: very young drivers (16 to 19) and very old drivers (80+) have significantly higher loss rates than middle-aged drivers. Driving record provides direct behavioral evidence, with at-fault accidents and moving violations strongly predicting future claims.

Vehicle-related factors include make, model, model year, vehicle use (commute, pleasure, business), annual mileage, and safety features. Sports cars and high-performance vehicles tend to have higher collision and comprehensive loss costs. Newer vehicles cost more to repair, increasing severity. Vehicles with advanced driver assistance systems (ADAS) may have lower collision frequency but higher repair costs when incidents do occur due to the expense of recalibrating sensors and cameras.

Credit-Based Insurance Scores

Credit-based insurance scores are among the most powerful predictive variables in personal auto insurance, though their use remains controversial. Actuarial studies have consistently shown a strong statistical relationship between credit history and insurance loss experience: individuals with better credit scores tend to file fewer claims and have lower average claim amounts. The reasons for this correlation are debated, with theories ranging from financial responsibility as a proxy for general risk-taking behavior to the stress and distraction associated with financial difficulties.

Most states permit the use of credit-based insurance scores in auto insurance pricing, but several states (including California, Hawaii, and Massachusetts) prohibit or severely restrict their use. Where permitted, insurance scores are typically computed using a formula that weighs payment history, outstanding debt, length of credit history, pursuit of new credit, and credit mix. Actuaries must ensure that the scoring model complies with state-specific regulations and that its use does not produce unfairly discriminatory outcomes for protected classes.

The regulatory landscape around credit scoring continues to evolve. Some states have implemented restrictions on how credit information can be used during policy renewals, prohibiting adverse actions based solely on credit changes. Actuaries working with credit-based scoring models must stay current with these regulatory developments and be prepared to adjust their pricing models accordingly.

Territorial Rating

Territory is among the most powerful rating variables in auto insurance because loss costs vary dramatically by geography. Urban areas have higher collision frequency due to traffic density, while rural areas may have higher severity due to higher speeds. Theft rates, litigation environments, medical cost levels, and weather patterns all vary by location. A ZIP code in downtown Detroit might have loss costs five times higher than a rural ZIP code in the same state.

Actuaries define territories using ZIP codes, counties, or custom geographic units and estimate territorial relativities using credibility-weighted combinations of local and statewide experience. The credibility-weighting process is essential because many geographic units have too little data to produce stable estimates on their own. A ZIP code with only 50 policies cannot produce a reliable loss cost estimate from its own experience, but it can be improved by blending with the experience of nearby areas.

Spatial smoothing techniques help stabilize estimates for areas with limited data while preserving meaningful geographic variation. Kriging, geographically weighted regression, and spatial autoregressive models allow the actuary to borrow information from neighboring areas in a statistically principled way. These techniques produce smoother territorial relativities that reflect genuine geographic patterns rather than random noise from small sample sizes.

Building and Validating the Rating Model

Actuaries fit generalized linear models (GLMs) to large datasets to estimate the independent effect of each rating variable on loss frequency and severity. The multiplicative structure of GLMs with a log link is particularly well suited to auto insurance pricing because the overall rate can be expressed as a base rate multiplied by a series of relativities, one for each rating variable. This structure aligns with how insurance rate manuals are organized and filed with regulators.

Separate models are typically built for each coverage (bodily injury liability, property damage liability, collision, comprehensive, personal injury protection, uninsured motorist) because the rating variables and their effects differ across coverages. For example, vehicle characteristics are more important for comprehensive coverage (where the peril is theft, vandalism, or weather damage) than for bodily injury liability (where the peril is an at-fault accident causing injury to others).

Model validation involves comparing predicted and actual loss ratios across each rating variable to confirm that the model captures the key patterns in the data. Lift charts assess the model's ability to discriminate between high-risk and low-risk policies. Double lift charts compare the proposed model against the current rating plan to quantify the improvement in risk segmentation. Out-of-sample testing (holding out a portion of the data or using a more recent time period) checks that the model generalizes beyond the training data.

Regulatory Considerations

Auto insurance pricing is heavily regulated at the state level. Most states require that rates be filed with and approved by the state insurance department before they can be used. The filing must demonstrate that the rates are not inadequate, not excessive, and not unfairly discriminatory. Actuaries must provide supporting documentation showing the data, methods, and assumptions used to develop the proposed rates.

Certain rating variables face specific regulatory restrictions. As noted, some states prohibit gender-based pricing. Others restrict or prohibit the use of education level, occupation, or ZIP code as rating factors due to concerns about proxy discrimination. The regulatory treatment of telematics data (driving behavior captured by devices or smartphone apps) is evolving, with states taking different approaches to how this data can be used in rating.

Actuaries must balance the predictive accuracy of their models with regulatory and social considerations. A variable may be highly predictive of loss costs but raise fairness concerns if it correlates strongly with race, income, or other protected characteristics. The ability to communicate clearly about why specific rating variables are used and how they relate to the underlying risk is essential for successful regulatory filings.

Emerging Trends in Auto Insurance Pricing

Several trends are reshaping auto insurance pricing. Telematics and usage-based insurance programs are enabling more granular, behavior-based pricing. Advanced driver assistance systems are changing the frequency and severity profiles of modern vehicles. The prospect of autonomous vehicles raises long-term questions about liability and pricing structures. Inflation in vehicle repair costs (driven by the complexity of modern vehicles and supply chain disruptions) is affecting severity trends across all coverages.

Machine learning techniques, particularly gradient boosting machines, are being used alongside GLMs to identify non-linear effects and interactions that traditional models miss. These techniques are most commonly used as diagnostic tools to improve GLM specifications rather than as standalone pricing models, because GLMs remain the standard for regulatory filings due to their interpretability. Actuaries who can effectively combine traditional and modern modeling techniques are well positioned to develop competitive, accurate, and regulatorily compliant rating plans.