Resumen de Joint modelling for customer lapses in the insurance sector
In this article, insurance customer lapsing is analized from a joint modelling perspective. Based on the ROC curve performance, it compares the predictive capacity of three churn prediction models: two univariate models, one for each line of business -auto and home-,and the bivariate model for the two different lines of business jointly. The estimation is based on a sample of real customers of a major insurance company in Spain. The results show that, if the information from both business lines simultaneously is combined, the predictive power of the bivariate model is enhanced. On the contrary, if only it is considered univariate models or marginal effects, then the predictive power decreases. The main contribution of this study is bringing to light the importance of taking into account global policyholder information from several products in order to improve decision making processes involving customers in the insurance sector.
