Implementation of penalized survival models in churn prediction of vehicle insurance

Abstract

The insurance industry plays an indispensable role in the financial structure of today’s world. Vehicle insurance accounts for a large part of the income of insurance companies, and retaining regular clients has proved economical, making car insurance clients’ churn management a priority. Meanwhile, clients’ churn will determine companies’ financial status and reflect the service and management level of the company. This paper proposes a new model, i.e., combined the Cox model with variable penalties (Lasso, SCAD, and MCP) to model the clients’ churn problems and determine the crucial factors that affect clients’ decisions, based on their personal information and behavior data, which are provided by a large insurance company in China. This model is proved to be successful in identifying the client churning and making comparisons among the penalties. The variable penalties model reveals the most important factor about clients’ churn problems, which can provide a reliable basis for the product development of insurance companies. MCP creates the greatest sparsity but consequently loses some information. We suggest using SCAD in the model, as it balances the sparsity and information reservation. Furthermore, a new approach is proposed to building a dynamic threshold of churning probabilities, which companies can use to manage their clients more flexibly. In this way, clients who are prone to churn can be identified in advance, and the maintenance cost of client management can be reduced accordingly.