Lyn Hunstad, Robert Bernstein, and Jerry Turem, 1994, "Impact Analysis of Weighting Auto Rating Factors to Comply with Proposition 103," California Department of Insurance, December.
Proposition 103 requires that the safety record, mileage, and driving experience rating factors have the greatest influence on auto premiums. However, there has never been any generally accepted procedure for measuring the weight of a rating factor. Since the weight of a rating factor basically determines how auto insurance will be priced for millions of consumers, this issue has central importance for implementing Proposition 103. The amount consumers pay for auto insurance will vary from current practices depending how the weight of a rating factor is measured. This variation, increase or decrease, is called premium variations. The analysis in this report estimates the impact of alternate approaches to meeting this requirement.
Rating factors are the means used to vary the cost of insurance to different drivers and different vehicles. Each rating factor is divided into two or more categories. Generally, the category into which one is assigned determines whether one will receive a discount or a surcharge. For example, the mileage rating factor could be divided into five categories: very low, low, average, high, and very high annual mileage. Typically the low and very low categories would receive discounts, the high and very high categories would be surcharged, and the average mileage category would see no change in the premium level. The amount of discounts and surcharges associated with a rating factor effect its ability to influence the final premium paid by consumers. The weight of a rating factor is a measurement of its influence on premium.
Various approaches to the problem of measuring the weight of a rating factor were explored in a June 1993 report by the Department, and by a technical symposium, held by the Department in January 1994, to explore potential methods. These efforts resulted in the identification of several possible methods for measuring weight. Under certain circumstances many of the methods had serious difficulties producing a reliable, meaningful measurement of weight. Eliminating the methods that would likely have implementation problems left two methods: the Single Omit, and the Average Class.
The Single Omit method calculates the weight of a rating factor by examining the effect on premium if the factor is omitted from the premium calculation process. The Average Class method calculates the weight of a rating factor by calculating the average differences of the discounts and surcharges associated with the factor. Key differences between the methods are:
- The Single Omit method weight reflects the average of all the individual insureds.
- The Single Omit method requires extensive calculations and detailed data on all individuals.
- The Average Class method weight is quick and easy to calculate.
- The Average Class method does not require detailed data on all individuals.
- The Average Class method may not reflect the average of all the individual experiences.
- To be effective, the Average Class method requires standardized factors, the Single Omit method does not.
In order to produce as accurate as possible an estimate of the effect of implementing either of these weighting methods, the Department created a large database to represent all of California's auto insurance consumers. This database contains detailed individual records from the top eleven auto insurer groups plus a major writer of sub-standard risks. In total, it includes over 11 million records with information on each individual's driving record, annual mileage, years licensed, vehicle characteristics, use, zipcode where garaged, coverage levels, premiums charged, and much more.
At the same time the Department built the database, it collected detailed information on the rating factors used by insurers. Every category used by every rating factor for every insurer and the rate associated with it was identified and programmed into a computer. After analyzing these rating factors wide variations were found in how different insurers created and used the same rating factors. These substantial variations from insurer to insurer appear to be arbitrary. For example, the mileage rating factor ranged from two to ten categories. One insurer with only two categories divided the mileage categories at 20,000 miles, while another divided them at 7,500.
Both the weighting methods were evaluated using consistent, standardized rating factors that were identical for each insurer. These standardized rating factors eliminated the arbitrary variations from insurer to insurer. If implemented in the market, they would act to encourage competition between insurers based on price and service, rather than avoidance of certain types of business. The Single Omit method was also evaluated without the standardized factors, using mostly insurers' current versions of the rating factors.
All insurers used some type of sequential analysis process when they developed their rates. Sequential analysis is a method for determining the rates (i.e., the amount of discount or surcharge) to associate with the categories of a rating factor. It does not produce any measurement of a factor's influence on premium. Therefore, it is not a weighting method and could not be included in this analysis.
A microsimulation procedure was used to evaluate the two weighting methods. This method is unique for auto insurance analysis although it is used extensively in other policy research areas. This method requires data on individuals rather aggregate or grouped data. It involves two basic steps. In the first step a model of the insurer's current practices is developed. In the second step new models were developed to reflect the two weighting methods by recasting the factors and formulas to meet Prop. 103 requirements. To estimate the impact of a weighting method, the estimated premium was compared to the individual's current premium from the model of current rating practices. A version of each model was built separately on a company by company basis, and the effect of the model on each individual at each of the insurers was determined. After measuring the model's effect for over 9 million consumers, the results were summarized to show the effect of different consumer groups and the different geographic areas in the state.
None of the insurers analyzed are currently complying with the requirements that auto premiums be primarily determined by the safety record, mileage, and driving experience rating factors in that rank order. The mileage rating factor needed the most modifications to come into compliance with Prop. 103, its weight had to be increased between 132% and 2,123% depending on the company and the model being evaluated. The failure of insurers' current rating practices to give enough influence to the number of miles driven annually is the single largest source of premium variations among all the rating factors.
For most consumer groups there is little average change in premium from what they currently pay. Most of the larger average changes occur at the extreme ends of the mileage groups (the very low mileage driver and the very high mileage drivers), and among the very young or inexperienced drivers. These changes seem to be consistent with the intent of Prop. 103. For both methods using the standardized factors: Los Angeles County averaged a reduction in premium of $7 to $8, while Sacramento and Fresno Counties averaged increases of $13 to $14, and the San Francisco bay area averaged around a $4 increase.
Each model was required to be revenue neutral. One consequence of this is that every dollar that is reduced from one consumer's premium must be added to the premium of another consumer. Overall, few consumers saw large increases or decreases. Most consumers are in the middle and experience either no change, a slight increase, or a slight decrease. The larger and larger increases are experienced by fewer and fewer consumers. Similarly, the larger and larger decreases are experienced by fewer and fewer consumers (see figure below). Only a few percent see an increase of over $100 and only a few percent see decreases greater than $100 in their six month bodily injury premium. When premium variations is classified as furthering the intent of the Proposition (positive) or not (negative), 90% of all consumers experienced either positive premium variation or a change in premium of 10% or less.
Because of the extensive amount of work that was required to duplicate each insurer's current rating practices, and model the new weighting methods, most analyses were limited to bodily injury coverage, the largest component (36%) of total auto insurance costs. For one large company, the Single Omit without standardized factors model was developed for all coverages. These results were compared to the model developed for only bodily injury and property damage coverages. The premium variations to the total premium for all coverages was about 20% less than expected based on bodily injury's percent of total premium. This was primarily because of less variability in premium among the non-liability coverages. This flatter rating structure makes achieving compliance with Prop. 103's weighting requirements easier, and less premium variations is produced when the coverages do come into compliance.
The report also discusses other issues related to implementing a weighting method and areas for future research. A basic issue is how precisely the Department should specify the procedures to be use in developing the rates associated with the rating factors. Related to this is the option of the Department collecting and making available data and analyses that could be used to develop rates.