In the state of Massachusetts, every driver is guaranteed automobile insurance at state-determined rates. Insurance companies cannot deny coverage to any driver, but the companies do have the option to cede a driver to a state agency called CAR if they expect to lose money on the policy. The company needs very good estimates for the expected loss on a policy in order to make the best possible cession decision. The main difficulty facing the insurance company is that the groups of drivers where the decision is critical are also the groups of drivers with the least reliable data. It is the most difficult to predict losses for exactly the policies that need the best predictions. This project describes two ways that a company can use loss data to obtain better predictions for individuals in groups with sparse data. The basic idea is to use a weighted average of "neighboring" policies to improve predictions for these borderline policies. The weights in the average are computed by using consecutive years in the company's loss data. One of the main results of this project is that there is no clear pattern leading to an easy and uniform choice of the weights.

• This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator

• Hogan, Elizabeth Anne.

Publisher

• Worcester Polytechnic Institute

Identifier

• 00E012M

Advisor

• Heinricher, Arthur C.

Year

• 2000

Date created

• 2000-01-01

Resource type

• Major Qualifying Project

Major

• Mathematical Sciences

Rights statement

• In Copyright