Value-enhancing modeling of surrenders and lapses

Abstract

Voluntary terminations of life insurance policies mean customer churns that usually lead to losses. Accurate predictions of voluntary terminations facilitate churn management, the valuation of life insurance policies, and the (asset-liability) management of life insurers. We use real-world data with adequate explanatory variables to evaluate the performance of three machine learning methods relative to the performance of three statistical methods in predicting voluntary terminations. Moreover, we decompose voluntary terminations into surrenders and lapses and find that some factors used to predict surrenders differ from those used to predict lapses. Then, we establish a two-stage model for insurers to take cost-effective actions to reduce the propensities of surrenders and lapses. This model outperforms conventional ones in terms of the resulting NPV (net present value).