Speaker(s): Salma Jamal (KPMG ), Lorenzo Invernizzi (Zurich Insurance Company Ltd.)
Actuaries are well acquainted with traditional reserving methods such as chain ladder and Bornhuetter-Ferguson. These traditional actuarial methods have been shown over many decades of actuarial practice, and in certain circumstances, to work well at an aggregate level to calculate appropriate reserve provisions. However, today with improvements in analytical methods and technology, actuaries are better equipped to uncover detailed loss drivers using Machine Learning techniques. Nowadays, insurers are highly interested in the anatomy of their portfolios. From policyholders’ behaviours, business strategies or portfolio pruning, sources of data volatility are numerous and impact both the reserving and the pricing processes of an insurer. Machine Learning methods can identify obscure trends in the data and incorporate suitable adjustments in its forecasts. When used together, traditional methods and machine learning methods can both support reserve estimates as well as provide the business explanations that are demanded by our stakeholders. The aim of our working group is to demonstrate that their joint application can be a powerful decision-making tool.Our study focused on a real data set provided by Swiss Re. The Line of Business concerned is Professional Liability. After a processing phase, traditional methods such as chain ladder and GLM have been compared to various Machine Learning methods. A synthesis completes the study, highlighting its limits and suggesting avenues of reflection for future research.