Lessons in Lapse Prediction: A Machine Learning Approach
A Talk by Patricia Wang MSc FIA FASSA and Jennifer Loftus FIA FSAI
About this Talk
Lapse rate is one of the main key risk drivers for many life and health insurance contracts and a key assumption for many actuarial exercises such as pricing, reserving, ALM, and risk management . Despite being influenced by many factors such as duration, contract types and entry age, lapse rates modelling has been rather crude in practice. We would like to share our findings on how different machine learning techniques can be applied to model lapse rates, and compare the performances of various machine learning techniques against more traditional approaches. The techniques we considered include: regression model, Bayesian mix, XGB and Neural Network. We further demonstrate how explainable AI can be applied to explain model predictions and gain further insights into the driving forces of lapse rates. By comparing these new techniques against the current standard models, we assess the benefits and challenges of adopting these new techniques.
