|
Accurate modelling of lapse rates in life insurance portfolios is a critical component of actuarial work, underpinning cash flow projections, embedded value calculations, and regulatory frameworks such as IFRS 17 and Solvency II. This talk explores lapse modelling techniques, focusing on their practical use in long-term liability projections.
The purpose of the session is to provide a comprehensive technical overview of methods used to model life insurance lapse and surrender rates, while demonstrating how these approaches integrate with actuarial cash flow models, which support accounting, pricing, risk, and solvency assessments.
Key determinants of lapse behaviour will be discussed, ranging from market-dependent factors such as interest rates, unemployment, and economic cycles to contract-specific influences like product design, policyholder demographics, and distribution channels. The session will also include a review of empirical findings, including competing hypotheses that link lapse rates to interest rate movements or policyholder liquidity needs.
Several modelling concepts will be overviewed. These include traditional segment-based assumptions, statistical approaches such as one-factor models, generalized linear models (GLM) with emphasis on logistic regression, and their extensions. Machine learning approaches such as classification and regression trees, random forests, gradient boosting, and neural networks will also be considered, with attention to their opportunities and constraints in actuarial contexts.
From a technical perspective, the session will focus on Key determinants of lapse behaviour, variable selection, and diagnostic tools, including residual analysis, goodness-of-fit tests, and predictive performance metrics. Backtesting and validation strategies will also be discussed. Finally, the session will address practical aspects of implementation, including deployment in R and/or Python, integration with actuarial systems and its challenges.
Early-bird discount is available for bookings made by 30 April 2026.
|
