Grouping of contracts in insurance using neural networks

Abstract

ABSTRACT Despite the high importance of grouping in practice, there exists little research on the respective topic. The present work presents a framework for grouping and a novel method to optimize model points in life insurance. We introduce a supervised clustering algorithm using neural networks to form a less complex portfolio, alias grouping. In a two-step approach, we first approximate selected characteristics of a portfolio. Next, we nest this estimator in a neural network, such that cluster representatives, alias model points, are calibrated in accordance with their effect on the characteristics of the portfolio. This approach is similar to the work by Horvath, B., Muguruza, A. & Tomas, M. [(2019). Deep learning volatility. Available on arXiv 1901.09647.], who focus on the calibration of implied volatility models. Our numerical experiments for term life insurance and defined contribution pension plans show significant improvements, in terms of capturing the characteristics of a portfolio, of the neural network approach over K-means clustering, a common baseline algorithm for grouping. These results are further confirmed by a sensitivity analysis of the investment surplus, where we additionally show the flexibility of the model to include common industry practice.