Practical use of machine learning in non-life pricing – smart price architecture

Practical use of machine learning in non-life pricing – smart price architecture

301 Views
Added in AFIR / ERM ASTIN / NON-LIFE LIFE

Thanks! Share it with your friends!

URL

You disliked this video. Thanks for the feedback!

Sorry, only registred users can create playlists.
URL

Description:

Speakers: Frank Schönfelder, Clemens Frey

New techniques related to machine learning, big data and analytics gain more and more attention from the actuarial community, but also decision-makers the insurance industry. There is also a trend in the German market of P&C companies setting up own data science groups focusing on implementing data science methods in order to enhance the business capabilities of their companies. In general, this refers to insurance claims management, to pricing, but also to advanced methods for reserving and risk management.

We have engaged in several projects in which machine learning techniques have been used in the insurance practise. We believe that some aspects of our work have the potential to set a new standard in the actuarial usage of machine learning within P&C pricing - the Smart Price Architecture.

Smart Price combines well-known GLM pricing techniques that with machine learning techniques. By adding machine learning risk-models on own data, the GLM risk model is being enhanced, without having to change the GLM or add complexity to the GLM. In addition, it is possible to reverse engineer existing price-models (and to a certain extend risk-models too) with machine learning and add those into the Smart Price. In addition, client-oriented scoring model may also be factored in. When combining those different risk-models, data science methods like stacking are used. The architecture also works in situations where there is a high influence from underwriting on the price setting.

We would like to give a top-down overview on the Smart Price Methodology, and present results we have gained with this kind of approaches. We will also show that, by applying this methodology, the gap between the desire of the insurance industry to use modern machine learning techniques and the application of existing, well-proven methodologies can be closed and thus the value creation of actuarial methods extended.

Post your comment

Sign in or sign up to post comments.
Be the first to comment