Application of a Bayesian Choice Model on Monthly Client Data - at SpareBank 1 SMN

Abstract

This paper applies a Bayesian Dynamic Choice model on monthly collected client data at SpareBank 1 SMN in order to examine the rate at which they enter or leave Life- and Danage-insurance.

The main goal of the paper is to develop a model with predictive abilities. To reflect this view, the data-set is divided into two part. The training-data is based on entries from November 2013 throughout December 2014. Additional test-data became available in 2015 and have entries from January 2015 to the end of February 2015.

A model will be presented that were able to identify approximately 25% of the observed activity of purchasing Life-insurance among a client group that of 0.1%. Additional 35% of the activity were identified among 12% of the clients, meaning that the model have a True Positive Rate of 0.6 and a False Positive Rate of 0.12. This far exceed the performance of a random guess strategy.

The predictive ability of identifying purchases of Damage-insurance is equally strong, while it is weaker for the identification of clients that leave the products.