CASE STUDY: P&C INSURANCE RATEMAKING

Context

• Large North American automobile insurer
• Annual volume of two billion dollars in direct written premiums

Problem

• To review the entire ratemaking currently in use, in particular market segmentation, in order to gear upcoming rate changes.
• Data base includes historical policy and claim records of the last seven years.
• Large fluctuations in claim frequency from one segment to another
• A small number of very costly claims render severity modeling a daunting task

Solution

• ApSTAT conducted an extensive study including:
  • Evaluating the impact of individual variables and pairs of variables on risk level in order to identify the most predictive variables or combinations or variables that could be added to the current set of ratemaking criteria.
  • Evaluating the out-of-sample performance of different data-mining models including neural networks, decision trees, and generalized linear models.
  • The development of a modelling architecture allowing to solve the specific risk assessment problem of automobile insurance, i.e., regression in the presence of asymmetric heavy-tailed noise.

Benefits

• Increased profitability of approx. 2% of volume using the recommended ratemaking

• Better understanding of the impact of different variables on the risk level associated to each insured
• Guidance of future internal research towards the suggested avenues
• Better understandng of the different data-mining models and their potential impact on automobile insurance ratemaking

To Know More

Read the white paper Data Mining Algorithms for Actuarial Ratemaking.

« Back to Case Studies