request info email to friend
New Auto Insurance Rating System
By David Miller
How much should an insurance company charge you for your insurance? Insurance companies have always used statistics to answer that question. For example, in auto insurance, it is reasonable to assume that younger drivers probably will have more accidents than older drivers, or that someone who drives 30 miles to work is more likely to have an accident than someone who drives two miles to work.
Until recently, most of this statistical information was used in a “cause and effect” fashion. During rate reviews, a company would use statistical “indications” to see if a particular state or territory needed more premium to offset poor claims experience or if a reduction in premium was warranted because of low claims costs. Other factors, such as the desire to increase or decrease market share or respond to the competition, also would be considered in pricing reviews. After the prices were implemented, the company would compare the actual results with the anticipated results in the next review, and the whole cycle would start over again.
Now, many insurance companies are implementing predictive modeling to determine their pricing. Consider the amount of data an auto insurance policy contains. Your declarations page has your name and address; the types of cars you drive; the names, dates of birth and driver’s license numbers of everyone in your home; and the amounts of coverage you purchased. The underwriting department has a five-year summary of any tickets you’ve received and every claim you’ve had – even those little $50 towing claims. The claims department has detailed notes on all of your claims, including the facts surrounding the claim and how much was paid for each type of coverage involved in the claim. Names, addresses, and phone numbers of witnesses, attorneys and body shops are also in the claim file.
What if all of these hundreds of variables could be accessed on every customer at the same time and distilled to 50 or 60 key attributes? Taking it a step further, what if these 50 to 60 key attributes could be compared against one another in every possible combination to find relevant correlations? Finally, what if these 50 to 60 attributes were assigned a mathematical score and combined into an equation or algorithm? The result would be a predictive model that charges the correct price for any given exposure.
One of the companies represented by Bensman Risk Management recently completed an overhaul of its auto insurance program and implemented a predictive modeling platform for pricing and underwriting. While creating this plan, the company also expanded the number of pricing tiers and territories to create millions of possible outcomes.
In the short run, this may make it more difficult for us to give you immediate answers about how things like having an accident will affect your rates, because the number of variables has increased so greatly. But in the long run, it should bring us much closer to the ultimate goal of insurance rating: finding a premium charge that reflects the actual risk as closely as possible.