Predictive modeling of lapse risk: An international financial services case study

Background. Stability and growth in life insurance market is an important economic indicator. Therefore, yearly coverage lapse-rate estimates are one of the key statistics that actuarial analysts need to characterize and manage the insurance business. Aim. We aim to present machine learning based approach applied to a real data set covering a decade long customer history of an international financial services corporation. Methodology. A challenging issue with this problem is evolving nature of the lapse determinants. We present diagnostic analysis characterizing the relationship among variables and their respective predictive value. Results. Particularly, a number of fixation variables are identified that cause over-fitting. Then a predictive modeling approach and validation results are presented to analyze the comparative strengths of different algorithms. Conclusion. Our results show that Random Forest performed better than logistic regression and naive Bayes in terms of specificity and sensitivity measures.