Predicting the Lapse Rates of AllSecur

The lapse rate of the clients of AllSecur is predicted by three different methodologies. The current method used by AllSecur is the Generalized Linear Model and serves as benchmark. The two new methodologies are survival analysis and machine learning. I select the best method by evaluating its predictive performance in an out-of-sample dataset. The best survival analysis method is the Cox Proportional Hazards model with variables selected by a Lasso regression. The best machine learning method is the Stochastic Gradient Boosting algorithm. I find that the Stochastic Gradient Boosting algorithm outperforms the GLM and the Lasso regression, and that the GLM outperforms the Lasso regression.