WEALTH AND ASSET MANAGEMENT THOUGHT LEADERSHIP

Neural Networks for Motor Insurance

White paper


Abstract

In the actuarial practice of motor insurance companies, estimating a fair price to charge customers for an insurance contract remains a challenging problem. 

The determination of the insurance premium is generally done by calculating the expected average claim for each customer profile on the basis of a number of variables that reflect the risk factors associated with that particular type of customer.

On the one hand, the premium should be high enough to guarantee the company a profit margin in accordance with the customer’s risk profile; on the other hand, it should also be low enough to keep the company competitive on the market. While the calculation of the average claim can be carried out on the basis of data that is known completely only by the company, in determining the competitiveness of the selected premium strategy it is very important to have some idea of the strategies being adopted by competitors, so as to be able to identify and exploit growth opportunities through fine-tuning of the company’s own premiums. 

Reverse engineering of a competitor’s premiums means being able to determine the structure underlying the premiums charged to customers, given a number of details of the profile risk factors. In this case, what are needed are the premiums set by the company’s competitors. 

Neural networks are suitable for application to both of these problems. 

Customer profiles are composed of a high number of variables, and the technique used must be able to handle them without incurring in overfitting. Neural networks benefit in general from a higher number of degrees of freedom and can work with very large amounts of input data without scaling the computation to an unmanageable size. The high non-linearity of neural networks allows them to capture subtle relationships between variables, thus increasing the explanatory power of the method.

In this study, we focused on determining the premium structure of a company given a number of customer profile variables and the premium charged. This information was obtained using the online profiler of two insurance companies and collecting the proposed premiums.