Application of classical reserving techniques alongside machine learning algorithms and big data

Speaker: Satraajeet Mukherjee
Insurance companies deal with a lot of structured and unstructured claims data in their daily activities. With the growing popularity of big data and machine learning algorithms in other industries, there is a call for insurance companies to explore machine learning for the purposes of individual level reserving. Some companies are starting to see value in using machine learning techniques for reserving purposes, but the extent of their application is very limited due to the trade-off between simplicity and accuracy.
The transition to the use of predictive modelling and machine learning for reserving will be very gradual, and in the initial years, predictive models are being used to validate the outputs of classical techniques. This paper illustrates how machine learning techniques such as clustering and ensemble tree models can be used to supplement and enhance classical reserving techniques, to improve the accuracy of reserves.Classical reserving models, machine learning models, and a hybrid of classical and machine learning models will be applied to a large database of bodily injury claims, with a focus on granular reserving. This paper also demonstrates the value of these hybrid models, and proposes a way forward to implement these hybrid models as the insurance industry seeks to derive additional value from big data.